Healthcare Predictive Analytics Using Machine

Learning and Deep Learning Techniques: A Survey

Mohammed Badawy (  [email protected] )
Cairo University
Nagy Ramadan
Cairo University
Hesham Ahmed Hefny
Cairo University

Research Article

Keywords: Healthcare Prediction, Artificial Intelligence (AI), Machine Learning (ML), Deep Learning (DL),
Medical Diagnosis

Posted Date: August 22nd, 2022

DOI: https://doi.org/10.21203/rs.3.rs-1885746/v2

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Page 1/47
Abstract
Aim
This paper aims to present a comprehensive survey of existing machine learning and deep learning
approaches utilized in healthcare prediction, as well as identify inherent obstacles to applying these
approaches in the healthcare prediction domain.

Background
Healthcare prediction has been a significant factor in saving human lives in recent years. In the domain
of healthcare, there is a rapid development of intelligent systems for analyzing complicated relationships
among data and transforming them into real information for use in the prediction process. Consequently,
artificial intelligence is rapidly transforming the healthcare industry. Thus comes the role of systems
depending on machine learning as well as deep learning in the creation of steps that diagnose and
predict diseases, whether from clinical data or based on images, that provide tremendous clinical support
by simulating human perception and can even diagnose diseases that are difficult to detect by human
intelligence.

Methods
The studies discussed in this paper have been presented in journals published by IEEE, Springer, and
Elsevier. Machine learning, deep learning, healthcare, surgery, cardiology, radiology, hepatology, and
nephrology are some of the terms used to search for these studies. The studies chosen for this survey are
concerned with the use of machine learning as well as deep learning algorithms in healthcare prediction.

Results
A total of 40 working papers were selected and the methodology for each paper was clarified.

Conclusion
This paper presents a comprehensive survey as well as the current challenges in healthcare prediction.
studies have shown that artificial intelligence plays a significant role in diseases diagnosing.

1. Background
Each day, human existence evolves, yet the health of each generation either improves or deteriorates.
There are always uncertainties in life. Occasionally encounter a large number of individuals with fatal

Page 2/47
health problems due to the late detection of diseases. Concerning the adult population, chronic liver
disease would affect more than 50 million individuals worldwide. However, if the sickness is diagnosed
early, it can be stopped. Disease prediction based on machine learning can be utilized to identify common
diseases at an earlier stage. Currently, health is a secondary concern, which has led to numerous
problems. Many patients cannot afford to see a doctor, and others are extremely busy and on a tight
schedule, yet ignoring recurring symptoms for an extended length of time can have significant health
repercussions [1].

A medical diagnosis is a form of problem-solving and is a crucial and significant issue in the actual
world. Illness diagnosis is the process of translating observational evidence into disease names. The
evidence comprises data received from evaluating a patient and substances generated from the patient;
illnesses are conceptual medical entities that detect anomalies in the observed evidence [2].

Diseases are a global issue, thus medical specialists and researchers are exerting their utmost efforts to
reduce disease-related mortality. In recent years, predictive analytic models are playing a pivotal role in
the medical profession as a result of the increasing volume of healthcare data from a wide range of
disparate and incompatible data sources. Nonetheless, processing, storing and analyzing the massive
amount of historical data and the constant inflow of streaming data created by healthcare services has
become an unprecedented challenge utilizing traditional database storage [3, 4, 5].

The concept of medical care is used to stress the organization and administration of curative care, which
is a subset of healthcare [6]. The ecology of medical care was first introduced by White in 1961. White
also proposed a framework for perceiving patterns of health concerning symptoms experienced in
particular populations of interest, along with individual's choices in getting medical treatment. In this
framework, it is able to calculate the proportion of the population who used medical services over a
specific time period. The "ecology of medical care" theory has become widely accepted in academic
circles over the past few decades [7].

Healthcare is the collective effort of society to ensure, provide, finance, and promote health. In the 20th
century, there was a significant shift toward the ideal of wellness and the prevention of sickness and
incapacity. The delivery of health care services entails organized public or private efforts to aid persons in
regaining health and preventing disease and impairment [8]. Healthcare can be described as standardized
rules that help evaluate actions or situations that affect decision-making [9].

Healthcare is a multidimensional system. The basic goal of healthcare is to diagnose and treat illnesses
or disabilities. A healthcare system key component are health experts (physicians or nurses), health
facilities (clinics, hospitals that provide medications and other diagnostic services), and a funding
institution to support the first two [10].

With the introduction of systems based on computers, the digitalization of all medical records and the
evaluation of clinical data in healthcare systems have come to be a widespread routine practice. The
phrase "electronic health records" was chosen by the Institute of Medicine, a division of the National
Page 3/47
Academies of Sciences, Engineering, and Medicine in 2003 to define the records that continued to
enhance the healthcare sector for benefit of both the patients and physicians. Electronic Health Records
(EHR) are "computerized medical records for patients that include all information in an individual's past,
present, or future which occur in an electronic system used to capture, store, retrieve, and link data
primarily to offer healthcare and health-related services," according to Murphy, Hanken, and Waters [10].

Daily, healthcare services produce an enormous amount of data, getting it increasingly complicated to
analyze and handle it using "conventional ways." Using machine learning and deep learning, this data
may be properly analyzed to generate actionable insights. In addition, genomics, medical data, social
media data, environmental data, and other data sources can be used to supplement healthcare data.
Figure 1 provides a visual picture of these data sources. The four key healthcare applications that can
benefit from machine learning are prognosis, diagnosis, therapy, and clinical workflow, as outlined in the
following section [11].

The long-term investment in developing novel technologies based on machine learning as well as deep
learning techniques to improve the health of individuals via the prediction of future events reflects the
increased interest in predictive analytics techniques to enhance healthcare. Clinical predictive models, as
they have been formerly referred to, assisted in the diagnosis of persons with an increased probability of
disease. These prediction algorithms are utilized to make clinical treatment decisions and counsel
patients based on some patient characteristics [12].

Artificial Intelligence (AI) is a scientific field that successfully integrates computer science and large
datasets to solve problems. It requires an understanding of computing to build tools and devices that
offer desired behavior [13]. Figure 2 depicts machine learning and deep learning as subsets of AI.

Medical personnel are usually facing new problems, changing tasks, and frequent interruptions as a
result of the system's dynamism and scalability. This variability often makes disease recognition a
secondary concern for medical experts. Moreover, clinical interpretation of medical data is a challenging
task from an epistemological point of view. This not only applies to professionals with extensive
experience but also representatives, such as young physician assistants, with varied or little experience.
The limited time available to medical personnel, the speedy progression of diseases, and the fluctuating
patient dynamics all the time make diagnosis a particularly complex process. However, a precise method
of diagnosis is critical to ensuring speedy treatment and thus ensuring patient safety [14].

1.1 Machine Learning

Machine learning (ML) is a subfield of AI that aims to develop predictive algorithms based on the idea
that machines should have the capability to access data and learn on their own [15]. ML utilizes
algorithms, methods, and processes to detect basic correlations within data and create descriptive and
predictive tools that process those correlations. ML is usually associated with data mining, pattern
recognition, and deep learning. Although there are no clear boundaries between these areas and they
often overlap, it is generally accepted that deep learning is a relatively new subfield of ML that uses
Page 4/47
extensive computational algorithms and large amounts of data to define complex relationships within
data. As shown in Fig. 3, ML algorithms can be divided into three categories: supervised learning,
unsupervised learning, and reinforcement learning [16].

1.1.1 Supervised Learning

Supervised learning is an ML model for investigating the input-output correlation information of a system
depending on a given set of training examples that are paired between the inputs and the outputs [17].
The model is trained with a labeled dataset. It matches how a student learns fundamental math from a
teacher. This kind of learning requires labelled data with predicted correct answers based on algorithm
output [18]. The most widely used supervised learning-based techniques include K-Nearest Neighbor,
Naive Bayes, Support Vector Machines, Decision Trees, Random Forests, and Logistic Regression.

A. Linear Regression

Linear regression is a statistical method commonly used in predictive investigations. It succeeds in

forecasting the dependent, output, variable (Y) based on the independent, input, variable (X). The
connection between X and Y is represented as shown in Eq. 1 assuming continuous, real, and numeric
parameters.

Y = mX + c. (1)

where m indicates the slope and c indicates the intercept. According to Eq. 1, the association between the
independent parameters (X) and the dependent parameters (Y) can be inferred. [19].

The advantage of linear regression is that it is straightforward to learn, and it is also easy to eliminate
overfitting through regularization. One drawback of linear regression is that it is not convenient when it is
applied to non-linear relationships. However, it is not recommended for most practical applications as it
greatly simplifies real-world problems [20]. The implementation tools utilized in Linear Regression are
Python, R, MATLAB, and Excel.

As shown in Fig. 4, observations are highlighted in red, and random deviations' result (shown in green)
from the basic relationship (shown in blue) between the independent variable (x) and the dependent
variable (y) [21].

B. Logistic Regression

Logistic regression, also known as the logistic model, investigates the correlation between a large number
of independent variables and a categorical dependent variable, and calculates the probability of an event
by fitting the data to a logistic curve [22]. Discrete mean values ​must be binary, i.e., have only two
outcomes: true or false, 0 or 1, yes or no, or either superscript or subscript. In logistic regression,
categorical variables have to be predicted and classification problems to be solved. Logistic regression
can be implemented utilizing various tools such as R, Python, Java, and MATLAB [19]. Logistic regression
Page 5/47
has many benefits, such as it shows the linear relationship between dependent and independent variables
with the best results. It is also simple to understand. On the other hand, it can only predict numerical
output, is not relevant to non-linear data, is sensitive to outliers [23].

C. Decision Tree

The Decision Tree (DT) is the most popular supervised learning methods used for classification. It
combines the values of attributes based on their order either ascending or descending [24]. As a tree-
based strategy, DT defines each path starting from the root by a data separating sequence until a
Boolean conclusion is attained at the leaf node [25–26]. DT is a hierarchical representation of knowledge
interactions that contains nodes and links. When relations are employed to classify, nodes reflect
purposes [27–28]. An example of DT is presented in Fig. 5.

DTs have various drawbacks, such as increased complexity with increasing nomenclature, small
modifications that may lead to a different architecture, and more processing time to train data [19]. The
implementation tools used in DT are Python (Scikit-Learn), R Studio, Orange, KNIME, and Weka [23].

D. Random Forest

Random Forest (RF) It is a basic and most widely utilized algorithm that produces correct results most of
the time. It may be utilized for classification and also regression. The program produces an ensemble of
DTs and blends them [29].

In the RF classifier, the higher the number of trees in the forest, the more accurate the results. So, the RF
has generated a collection of DTs called the forest and combined them to achieve more accurate
prediction results. In RF, each DT is built only on a part of the given dataset and trained on
approximations. The RF brings together several DTs to reach the optimal decision [19].

As indicated in Fig. 6. RF randomly selects a subset of features from the data and from each subset it
generates n number of random trees [21]. RF will combine results from all DTs and provide them in the
final output.

Two parameters are being used for tuning RF models: mtry - the count of randomly selected features to
be considered in each division; and ntree - the model trees count. The mtry parameter has a trade-off:
large values ​raise the correlation between trees but enhance the per-tree accuracy [30].

The RF works with a labeled dataset to do predictions and build a model. The final model is utilized to
classify unlabeled data. The model integrates the concept of ​bagging with a random selection of traits to
build variance-controlled DTs [31].

RF offers significant benefits. First, it can be utilized for determining the relevance of the variables in a
regression and classification task [32, 33]. This relevance is measured with a scale, based on the impurity
drop at each node used for data segmentation [34]. Second, it automates missing values ​contained in the

Page 6/47
data and resolves the overfitting problem of DT. Finally, RF can efficiently handle huge data sets. On the
other side, RF suffers from drawbacks such as it needs more computing and resources to generate the
output results and it requires training effort due to the multiple DTs involved in it. The implementation
tools used in RF are Python Scikit-Learn and R [19].

E. Support Vector Machine

The most popular supervised ML algorithm for classification issues and regression models is called
Support Vector Machine (SVM). SVM is a linear model that offers solutions to issues that are both linear
and nonlinear. as shown in Fig. 7. Its foundation is the idea of margin calculation. The dataset is divided
into several groups to build relations between them [19].

SVM is a statistical-based learning method that follows the principle of structural risk minimization and
aims to locate decision bounds, also known as hyperplanes, that can optimally separate classes by
finding a hyperplane in a usable N-dimensional space that explicitly classifies data points. [35, 36, 37].
SVM indicates the decision boundary between two classes by defining the value of each data point, in
particular the support vector points placed on the boundary between the respective classes [38].

SVM has several advantages such as it works perfectly even with both semi-structured and unstructured
data. Kernel trick is a strength point of SVM. Moreover, it can handle any complex problem with the right
functionality and can also handle high-dimensional data. Furthermore, SVM generalization has less
allocation risk. On the other hand, SVM has many downsides. Its model training time is increased on a
large dataset. Choosing the right kernel function is also a difficult process. In addition, it is not working
well with noisy data. Implementation tools used in SVM include SVMlight with C, LibSVM with Python,
MATLAB or Ruby, SAS, Kernlab, Scikit-Learn, and Weka [23].

F. K - Nearest Neighbor

K-nearest neighbor (KNN) is an "instance-based learning" or non-generalized learning, which is often

known as a “lazy learning” algorithm [39]. KNN is used for solving the classification problems. To
anticipate the target label of the novel test data, KNN determines the distance of the nearest training data
class labels with a new test data point in the existence of a K value, as shown in Fig. 8. It then calculates
the number of nearest data points using the K value and terminates the label of the new test data class.
To determine the number of nearest-distance training data points, KNN usually sets the value of K among
0 and 1 [23].

KNN has many benefits such as it is sufficiently powerful if the size of training data is large. It is also
simple and flexible with attributes and distance functions. Moreover, it can handle multi-class data sets.
KNN has many drawbacks such as the difficulty of choosing the appropriate K value, it is very tedious to
choose the distance function type for a particular dataset, and the computation cost being a little high
due to the distance between all the training data points [31]. The implementation tools used in KNN are
Python (Scikit-Learn), WEKA, R, KNIME, and Orange [23].

Page 7/47
G. Naïve Bayes

Naive Bayes (NB) focuses on the probabilistic model of Bayes' theorem and is simple to set up as the
complex recursive parameter estimation is basically none, making it suitable for huge data sets [40]. NB
determines the class membership degree based on a given class designation [41]. It scans the data once
and thus classification is easy [42]. Simply, the NB classifier assumes that there is no relation between
the presence of a particular feature in a class and the presence of any other characteristic. It is mainly
targeted at the text classification industry [43].

NB has great benefits such as ease of implementation, can provide a good result even using fewer
training data, can manage both continuous and discrete data, ideal to solve prediction of multiclass
problems, and the irrelevant feature does not affect the prediction. NB, on the other hand, has the
following drawbacks: it assumes that all features are independent which is not always viable in real-
world problems, suffers the zero frequency problems, and the prediction of NB is not usually accurate.
Implementation Tools: WEKA, Python, R Studio, and Mahout [19].

1.1.2 Unsupervised learning

Unlike supervised learning, there are no correct answers and no teachers in unsupervised learning [43]. It
follows the concept that a machine can learn to understand complex processes and patterns on its own
without external guidance. This approach is particularly useful in cases where experts have no knowledge
of what to look for in the data and the data itself does not include the objectives. The machine predicts
the outcome based on past experiences and learns to predict the real-valued outcome from the
information previously provided, as shown in Fig. 9.

Unsupervised learning is widely used in the processing of multimedia content, as clustering and
partitioning of data in the lack of class labels is often a requirement [44]. Some of the most popular
unsupervised learning-based approaches are k-means, Principal Component Analysis (PCA), and Apriori
Algorithm.

A. k-means

The k means algorithm is the common portioning method [45] and one of the most popular unsupervised
learning algorithms that deal with the well-known clustering problem. The procedure classifies a
particular data set by a certain number of preselected (assuming k-sets) clusters [46]. The Pseudocode of
the K-means algorithm is shown in Pseudocode 1.

Page 8/47
 Pseudocode 1: k-means Pseudocode

1. Arrange K points in the space represented by the

clustered items. These points reflect the

centroids of the first group.

2. Set each object of the group that has the nearest

centroid.
3. After setting all the elements, the coordinates

of the k centroids have to be recalculated.

4. Repeat Steps 2 and 3 until the centroids stop

moving.

K means have several benefits such as being more computationally efficient than hierarchical grouping in
case of large variables. It provides more compact clusters than hierarchical ones when small k is used.
Also, the ease of implementation and comprehension of assembly results is another benefit. However, K-
Means have disadvantages such as the difficulty of predicting the value of K. Also, as different starting
sections lead to various final combinations, the performance is affected. It is accurate for raw points and
local optimization, and there is no single solution for a given K value - so the average of the K value must
be run multiple times (20–100 times) and then pick the results with the minimum J [20].

B. Principal Component Analysis

In modern data analysis, Principal component analysis (PCA) is an essential tool as it provides a guide
for extracting the most important information from a dataset, compressing the data size by keeping only
those important features without losing much information, and simplifying the description of a data set
[47, 48].

PCA is frequently used to reduce data dimensions before applying classification models. Moreover,
unsupervised methods, such as dimensionality reduction or clustering algorithms, are commonly used for
data visualizations, detection of common trends or behaviors, and decreasing the data quantity to name
a few only [49].

PCA converts the 2D data into 1D data. This is done by changing the set of variables into new variables
known as principal components (PC) which are orthogonal [24]. In PCA data dimensions are reduced to
make calculations faster and easier. To illustrate how PCA works, let's consider an example of 2D data.
When this data is plotted on a graph, it will take two axes. Applying PCA the data turns into 1D. This
process is illustrated in Fig. 10 [50].

C. Apriori

Page 9/47
Apriori algorithm is considered an important algorithm, that was first introduced by R. Agrawal and R.
Srikant, and published in [51, 52].

The principle of the Apriori algorithm is to represent the filter generation strategy. It creates a filter element
set (k + 1) based on the repeated k element groups. Apriori uses an iterative strategy called planar search,
where k item sets are employed to explore (k + 1) item sets. First, the set of repeating 1-items is produced
by scanning the dataset to collect the number for each item, then collecting items that meet the minimum
support. The resulting group is called L1. Then L1 is used to find L2, the recursively set of two elements is
used to find L3, and so on until no repeated k element groups are found. Finding every Lk needs a full
dataset scan. To improve production efficiency at the level-wise of repeated element groups, a key
property called the Apriori property is used to reduce the search space. Apriori property states that all non-
empty subsets of a recursive element group must be iterative. A two-step technique is used to identify
groups of common elements: join and prune activities [53].

Although it is simple, the Apriori algorithm suffers from several drawbacks. The main limitation is the
costly wasted time to contain a large number of candidate sets with a lot of redundant item sets. It also
suffers from low minimum support or large item sets and multiple rounds of data are needed for data
mining which usually results in irrelevant items, in addition to difficulties in discovering individual
elements of events [54, 55].

1.1.3 Reinforcement learning

Reinforcement learning (RL) is different supervised learning and unsupervised learning. It is a goal-
oriented learning approach. RL is closely related to an agent (controller) that takes the responsibility for
the learning process to achieve a goal. The agent, in particular, chooses actions, and as a result, the
environment changes its state and returns rewards. Positive or negative numerical values are used as
rewards. An agent's goal is to maximize the rewards accumulated over time. A job is a complete
environment specification that identifies how to generate rewards [56]. Some of the most popular
reinforcement learning-based algorithms are the Q-Learning algorithm and the Monte-Carlo Tree Search
(MCTS).

A. Q-Learning

Q-Learning is a type of model-free RL. It can be considered an asynchronous dynamic programming

approach. It enables agents to learn how to operate optimally in Markovian domains by exploring the
effects of actions, without the need to generate domain maps [57]. It represented an incremental method
of dynamic programming that imposed low computing requirements. It works through the successive
improvement of the assessment of individual activity quality in particular states [58, 59].

In information theory, Q-learning is strongly employed, and other related investigations are underway [60].
Recently, Q-learning combined with information theory has been employed in different disciplines such as
Natural Language Processing (NLP), pattern recognition, anomaly detection, and image classification [61,
62, 63, 64]. Moreover, a framework has been created to provide a satisfying response based on the user’s
Page 10/47
utterance using RL in a voice interaction system [65]. Furthermore, a high-resolution deep learning-based
prediction system for local rainfall has been constructed [66].

The advantage of developmental Q-learning is that it is possible to identify the reward value effectively
on a given multi-agent environment method as agents in ant Q-learning are interacting with each other.
The problem with Q-learning is that its output can stuck in the local minimum as agents just take the
shortest path [67].

B. Monte Carlo Tree Search

Monte Carlo Tree Search (MCTS) is an effective technique for solving sequential selection problems. Its
strategy is based on a smart tree search that balances exploration and exploitation. MCTS presents
random samples in the form of simulations and keeps activity statistics for better-educated choices in
each future iteration. MCTS is a decision-making algorithm that is employed in searching trees-like huge
complex regions. In such trees, each node refers to a state, which is also referred to as problem
configuration, while edges represent transitions from one state to another [68].

The MCTS is related directly to cases that can be represented by a Markov decision process (MDP), which
is a type of discrete-time random control process. Some modifications of the MCTS make it possible to
apply it to Partially Observable Markov Decision Processes (POMDP) ​[69]. Recently, MCTS coupled with
deep RL became the base of AlphaGo developed by Google DeepMind and documented in [70]. The basic
MCTS method is conceptually simple, as shown in Fig. 11.

Tree 1 is constructed progressively and unevenly. The tree policy is utilized to get the critical node of the
current tree for each iteration of the method. The tree strategy seeks to strike a balance between
exploration and exploitation concerns. Then, from the specified node, simulation 2 is run, and the search
tree is then updated according to the obtained results. This comprises adding a child node that matches
the specified node's activity and updating its ancestor's statistics. During this simulation, movements are
performed based on some default policy, which in its simplest case is to make uniform random
movements. The benefit of MCTS is that there is no need to evaluate the values of the intermediate state,
which significantly minimizes the amount of required knowledge in the field [72].

1.2 Deep Learning

Over the past decades, ML has had a significant impact on our daily lives with examples including
efficient computer vision, web search, and recognition of optical characters. In addition, by applying ML
approaches, AI at the human level has also been improved [73, 74, 75]. However, when it comes to the
mechanisms of human information processing (such as sound and vision), the performance of
traditional ML algorithms is far from satisfactory. The idea of ​D eep Learning (DL) was formed in the late
20th inspired by the deep hierarchical structures of human voice recognition and production systems. DL
breaks have been introduced in 2006 when Hinton built a deep structured learning architecture called
Deep Belief Network (DBN) [76].

Page 11/47
The performance of classifiers using DL has been extensively improved with an increased amount of
data compared to classical learning methods. Figure 12 shows the performance of classic ML algorithms
and DL methods [77]. The performance of typical ML algorithms becomes stable when they reach the
training data threshold, but DL upturns their performance as the amount of data increases [78].

DL (deep ML, or deep structured learning) is a subset of ML which involves a collection of algorithms
attempting to represent high-level abstractions for data through a model that has complicated structures
or otherwise, composed of numerous non-linear transformations. The most important characteristic of
DL is the depth of the network. Another essential aspect of DL is the ability to replace handcrafted
features generated by efficient algorithms for unsupervised or semi-supervised feature learning and
hierarchical feature extraction [79].

DL has significantly advanced the latest technologies in a variety of applications, including machine
translation, speech, and visual object recognition, NLP, and text automation, through the use of multi-layer
Artificial Neural Networks (ANNs) [16].

Different DL designs in the past two decades give the enormous potential for employment in various
sectors such as automatic voice recognition, computer vision, NLP, and bioinformatics. This section
discusses the most common architectures of DL such as Convolutional Neural Networks (CNNs), Long
Short-Term Memory (LSTM), and Recurrent Convolution Neural networks (RCNNs) [80].

A. Convolutional Neural Network

CNNs are special types of neural networks inspired by the human visual cortex and used in computer
vision. It is an automatic feed-forward neural network in which information transfers exclusively in the
forward direction [81]. CNN is frequently applied in face recognition, human organ localization, text
analysis, and biological image recognition [82].

Since CNN was first created in 1989, it has done well in disease diagnosis over the past three decades
[83]. Figure 13 depicts the general architecture of a CNN composed of feature extractors and a classifier.
Each layer of the network accepts the output of the previous layer as input and passes it on to the next
layer in feature extraction layers. A typical CNN architecture consists of three types of layers: convolution,
pooling, and classification. There are two types of layers at the network's low and middle levels:
convolutional layers and pooling layers. Even-numbered layers are used for convolutions, while odd-
numbered layers are used for pooling operations. The convolution and pooling layers' output nodes are
categorized in a two-dimensional plane called feature mapping. Each layer level is typically generated by
combining one or more previous layers [84].

CNN has a lot of benefits, including a human optical processing system, greatly improved 2D and 3D
image processing structure, and is effective in learning and extracting abstract information from 2D
information. The max-pooling layer in CNN is efficient in absorbing shape anisotropy. Furthermore, they
are constructed from sparse connections with paired weights and contain far fewer parameters than a

Page 12/47
fully connected network of equal size. CNNs are trained using a gradient-based learning algorithm and
are less susceptible to the diminishing gradient problem because the gradient-based approach trains the
entire network to directly reduce the error criterion, allowing CNNs to provide highly optimized weights
[84].

B. Long Short Term Memory

LSTM is a special type of Recurrent Neural Networks (RNNs) with internal memory and multiplicative
gates. Since the original LSTM introduction in 1997 by Sepp Hochrieiter and Jürgen Schmidhuber, a
variety of LSTM cell configurations have been described [92].

LSTM has contributed to the development of well-known software such as Alexa, Siri, Cortana, Google
Translate, and Google voice assistant [93]. LSTM is an implementation of RNN with a special connection
between nodes. The special components within the LSTM unit include the input, output, and forget gates.
Figure 14 depicts a single LSTM cell.

where

xt = Input vector at the time t.

ht-1 = Previous Hidden state.

ct-1 = Previous Memory state.

ht = Current Hidden state.

ct = Current Memory state.

[x] = Multiplication operation.

[+] = Addition operation.

LSTM is an RNN module that handles gradient loss problems. In general, RNN uses LSTM to eliminate
propagation errors. This allows the RNN to learn over multiple time steps. LSTM is characterized by cells
that hold information outside the recurring network. This cell enables the RNN to learn over many time
steps. The basic principle of LSTMs is the state of the cell, which contains information outside the
recurrent network. A cell is similar to a memory in a computer, which decides when data should be stored,
written, read, or erased via the LSTM gateway [94]. Many network architectures use LSTM such as
bidirectional LSTM, hierarchical and attention-based LSTM, convolutional LSTM, autoencoder LSTM,
network LSTM, cross-modal, and relational LSTM [95].

Bidirectional LSTM networks move the state vector forward and backward in both directions. This implies
that dependencies must be taken into account in both temporal directions. As a result of inverse state
propagation, the expected future correlations can be included in the network's current output [96].

Page 13/47
investigates and analyses this because bidirectional LSTM networks encapsulate spatially and
temporally scattered information and can tolerate incomplete inputs via a flexible cell-state vector
propagation communication mechanism. Based on the detected gaps in data, this filtering mechanism
reidentifies the connections between cells for each data sequence. Figure 15 depicts the architecture. A
bidirectional network is used in this study to process properties from multiple dimensions into a parallel
and integrated architecture [95].

Hierarchical LSTM networks solve multidimensional problems by breaking them down into sub-problems
and organizing them in a hierarchical structure. This has the advantage of focusing on a single or
multiple sub-problems. This is accomplished by adjusting the weights within the network in order to
generate a certain level of interest [95]. A weighting-based attention mechanism that analyses and filters
input sequences is also used in hierarchical LSTM networks for long-term dependency prediction [97].

Convolutional LSTM reduces and filters input data collected over a longer period of time using
convolution operations applied in LSTM networks or the LSTM cell architecture directly. Furthermore, due
to their distinct characteristics, convolutional LSTM networks are useful for modelling many quantities
such as spatially and temporally distributed relationships. However, many quantities can be expected
collectively in terms of reduced feature representation. Decoding or decoherence layers are required to
predict different output quantities not as features but based on their parent units [95].

The LSTM autoencoder solves the problem of predicting high-dimensional parameters by shrinking and
expanding the network [98].The autoencoder architecture is separately trained with the aim of accurate
reconstruction of the input data as reported in [99]. Only the encoder is used during testing and
commissioning to extract the low-dimensional properties that are transmitted to the LSTM. The LSTM
was extended to multimodal prediction using this strategy. To compress the input data and cell states,
the encoder and decoder are directly integrated into the LSTM cell architecture. This combined reduction
improves the flow of information in the cell and results in an improved cell state update mechanism for
both short-term and long-term dependency [95].

Grid Long Short-Term Memory is a network of LSTM cells organized into a multidimensional grid that can
be applied to sequences, vectors, or higher dimensional data like images [100]. Grid LSTM has
connections to eg the spatial or temporal dimensions of input sequences. Thus, connections of different
dimensions within cells extend the normal flow of information. As a result, Grid LSTM is appropriate for
the parallel prediction of several output quantities that may be independent, linear, or non-linear. The
network's dimensions and structure are influenced by the nature of the input data and the goal of the
prediction [101].

A novel method for the collaborative prediction of numerous quantities is the cross-modal and
associative LSTM. It uses a number of standard LSTMs to separately model different quantities. To
calculate the dependencies of the quantities, these LSTM streams communicate with one another via
recursive connections. The chosen layers' outputs are added as new inputs to the layers before and after
them in other streams. Consequently, a multimodal forecast can be made. The benefit of this approach is
Page 14/47
that the correlation vectors that are produced have the same dimensions as the input vectors. As a result,
neither the parameter space nor the computation time increase [102].

C. Recurrent Convolution Neural Network

CNN is a key method for handling various computers vision challenges. In recent years, a new generation
of CNNs has been developed, the Recurrent Convolution Neural Network (RCNN), which is inspired by
large-scale recurrent connections in the visual systems of animals. The Recurrent Convolutional Layer
(RCL) is the main feature of RCNN, which integrates repetitive connections among neurons in the normal
convolutional layer. With the increase in the number of repetitive computations, the Receptive Domains
(RFs) of neurons in the RCL expand infinitely, which is contrary to biological facts [103].

The RCNN prototype was proposed by Ming Liang & Xiaolin Hu [104, 105], the structure is illustrated in
Fig. 16, in which both forward and redundant connections have local connectivity and weights shared
between distinct sites. This design is quite similar to the Recurrent Multi-Layer Perceptron (RMLP)
concept which is often used for dynamic control [106, 107] (Fig. 17, middle). Similar to the distinction
between MLP and CNN, the primary distinction is that in RMLP, common local connections are used in
place of full connections. For this reason, the proposed model is known as RCNN [108].

The main unit of RCNN is the RCL. RCLs develop through discrete time steps. RCNN offers three basic
advantages. First, it allows each unit to accommodate background information in an arbitrarily wide area
in the current layer. Second, recursive connections improve the depth of the network while keeping the
number of mutable parameters constant through weight sharing. This is consistent with the trend of
modern CNN architecture to grow deeper with a relatively limited number of parameters. The third aspect
of RCNN is the time exposed in RCNN which is a CNN with many paths between the input layer and the
output layer, which makes learning simple. On one hand, having longer paths makes it possible for the
model to learn very complex features. On the other hand, having shorter paths may improve the inverse
gradient during training [103].

The primary goals of this work are to present a comprehensive overview of the key machine learning as
well as deep learning techniques employed in healthcare prediction, as well as to identify the obstacles
that machine learning and deep learning face in healthcare prediction.

The rest of this paper is structured as follows:

• Section 2 presents a survey methodology

• Section 3 gives a literature survey of the machine learning and deep learning techniques used in
healthcare prediction.

• Section 4 summarizes the advantage and limitations of the techniques discussed in section 3.

• Finally, Section 6 outlines the conclusions.

Page 15/47
2. Survey Methodology
The studies discussed in this paper have been presented and published in high-quality journals and
international conferences published by IEEE, Springer, and Elsevier. Machine learning, deep learning,
healthcare, surgery, cardiology, radiology, hepatology, and nephrology are some of the terms used to
search for these studies. The studies chosen for this survey are concerned with the use of machine
learning as well as deep learning algorithms in healthcare prediction. For this survey, empirical and review
articles on the topics were considered.

2.1 Survey Structure

This section discusses existing research efforts that healthcare prediction using various techniques in ML
and DL. This survey gives a detailed discussion about the methods and algorithms which are used for
predictions, performance metrics, and tools of their model.

2.1.1 ML-based Healthcare Prediction

In [109], the authors utilized a framework to create and assess ML classification models such as Logistic
Regression, KNN, SVM, and RF for the prediction of diabetes patients. ML method was implemented on
the Pima Indian Diabetes Database (PIDD) which has 768 rows and 9 columns. The forecast accuracy
delivers 83 percent accuracy. Results of the implementation approach indicate how the Logistic
Regression outperformed other algorithms of ML. The results indicated that only a structured dataset
was selected but unstructured data are not considered, also model should be implemented in other
healthcare domains like heart disease, and COVID-19, finally other factors should be considered for
diabetes prediction, like family history of diabetes, smoking habits, and physical inactivity.

In [110], The authors developed a diagnosis system focusing on 4 prediction algorithm models (RF, SVM,
NB, DT) to predict diabetes using two various databases (Frankfurt Hospital in Germany and PIDD
provided by the UCI ML repository). the SVM algorithm performed with an accuracy of 83.1 percent. There
are some aspects of this study that need to be improved, such as using a DL approach to predict diabetes
may lead to achieving better results, furthermore, the model should be tested in other healthcare domains
such as heart disease and COVID-19 prediction.

In [111], the authors proposed three ML methods (Logistic Regression - DT - Boosted RF) to assess the
COVID-19 OpenData Resources from Mexico and Brazil. To predict rescue and death, the proposed model
incorporates just the COVID-19 patient's geographical, social, and economic conditions, as well as clinical
risk factors, medical reports, and demographic data. On the dataset utilized, the model for Mexico has a
93 percent accuracy, and an F1 score is 0.79. On the other hand, on the used dataset, the Brazil model
has a 69 percent accuracy and an F1 score is 0.75. The three ML algorithms have been examined and the
acquired results showed that Logistic Regression is the best way of processing data. The authors should
be concerned about the usage of authentication and privacy management of the created data.

Page 16/47
In [112], The authors introduced a new model for predicting type 2 diabetes utilizing a network approach
as well as ML techniques (Logistic Regression, SVM, NB, KNN, Decision Tree, RF, XGBoost, and ANN). To
predict the risk of type 2 diabetes, the healthcare data of 1,028 type 2 diabetes patients and 1,028 non-
type 2 diabetes patients were extracted from de-identified data. The experimental findings reveal the
models’ effectiveness with an Area Under Curve (AUC) varied from 0.79 to 0.91. The RF model achieved
higher accuracy than others. This study relies only on the dataset providing hospital admission and
discharges summaries from one insurance company. External hospital visits and information from other
insurance companies are missing for people with many insurance providers.

In [113], The author proposed a healthcare management system that patients could use to schedule
appointments with doctors and verify their prescriptions. It gives support for ML to detect ailments and
determine medicines. ML models including DT, RF, logistic regression, and NB classifiers are applied to
the datasets of diabetes, heart disease, chronic kidney disease, and liver. The results showed that among
all the other models, logistic regression had the highest accuracy of 98.5 percent in the heart dataset.
while the least accuracy is of the DT classifier which came out to be 92 percent. In the liver dataset the
logistic regression with maximum accuracy of 75.17 percent among all others. In the chronic renal
disease dataset, the logistic regression, RF, and Gaussian NB, all performed well with an accuracy of 1. In
the diabetes dataset random forest with maximum accuracy of 83.67 percent. The authors should
include a hospital directory as then various hospitals and clinics can be accessed through a single portal.
Additionally, image datasets should be included to allow image processing of reports and the deployment
of DL to detect diseases.

In [114], the authors developed an ML model to predict the occurrence of Type 2 Diabetes in the following
year (Y + 1) using factors in the present year (Y). Between 2013 and 2018, the dataset was obtained as an
electronic health record from a private medical institute. authors applied logistic regression, RF, SVM,
XGBoost, and ensemble ML algorithms to predict the outcome of non-diabetic, prediabetes, and diabetes.
Feature selection was applied to choose the three classes efficiently. FPG, HbA1c, triglycerides, BMI,
gamma-GTP, gender, age, uric acid, smoking, drinking, physical activity, and family history were among
the features selected. According to the experimental results, the maximum accuracy was 73 percent from
RF, while the lowest was 71 percent from the logistic regression model. The authors presented a model
that used only one dataset. As a result, additional data sources should be applied to verify the models
developed in this study.

In [115], the authors categorized the diabetes dataset using SVM and NB algorithms coupled with feature
selection for enhancing the accuracies of the model. PIDD is taken from the UCI Repository for analysis.
For training and testing purposes the authors employed the K-fold cross-validation model, the SVM
classifier was performing better than the NB method it offers around 91 percent correct predictions,
however, the authors acknowledge that they need to extend to the latest dataset that will contain
additional attributes and rows.

Page 17/47
In [116], the authors introduced an unsupervised ML algorithm K-means clustering for the UCI heart
disease dataset to detect heart disease in the early stage. PCA is used for dimensionality reduction. The
outcome of the method demonstrates early cardiac disease prediction with 94.06 percent accuracy. The
authors should apply the proposed technique using more than one algorithm and use more than one
dataset.

In [117], the authors constructed a predictive model for the classification of diabetes data using the
logistic regression classification technique. the dataset includes 459 patients for training data and 128
cases for testing data. The prediction accuracy using logistic regression was obtained at 92 percent. The
main limitation of this research is that the authors have not compared the model with other diabetes
prediction algorithms and so it cannot be confirmed.

In [118], the authors developed a prediction model that analyses the user's symptoms and predicts the
disease using ML algorithms (DT classifier, RF classifier, and NB classifier) to solve health-related
problems by allowing professionals to predict diseases at an early stage. A dataset is a sample of 4920
patient records with 41 illnesses diagnosed. A total of 41 disorders were included as a dependent
variable. All of the algorithms achieved the same accuracy score of 95.12%. The authors noticed that
overfitting occurred when all 132 symptoms from the original dataset were assessed instead of 95
symptoms. i.e., the tree appears to remember the dataset provided and thus fails to classify new data. As
a result, just 95 symptoms were assessed during the data-cleansing process, with the best ones being
chosen.

In [119], the authors built a decision-making system that assists practitioners to anticipate cardiac
problems in exact classification through a simpler method and will deliver automated predictions about
the condition of the patient’s heart. implemented 4 algorithms (KNN, RF, DT, and NB), all these algorithms
were used in the Cleveland Heart Disease dataset. The accuracy varies for different classification
methods. The maximum accuracy is given when they utilized the KNN algorithm with the Correlation
factor which is almost 94 percent. The authors should extend the presented technique to leverage more
than one dataset and forecast different diseases.

In [120], the authors applied three classification methods (NB, SVM, DT, and KNN) to the Cleveland
dataset consisting of 303 cases and 76 attributes. Of these 76 traits, only 14 attributes are chosen for
testing. authors performed data preprocessing to remove noisy data. The KNN obtained the greatest
accuracy with 90.79 percent. To improve the accuracy of early heart disease prediction, the authors need
to use more sophisticated models.

In [121], the authors proposed a model to predict heart disease utilizing a cardiovascular dataset used in
this model and classified by using supervised ML algorithms (DT, NB, Logistic Regression, RF, SVM, and
KNN). The results reveal that the DT classification model predicted cardiovascular disorders better than
other algorithms with an accuracy of 73 percent. the authors highlighted that the ensemble ML
techniques employing the CVD dataset can generate a better illness prediction model.

Page 18/47
In [122], the authors attempted to increase the accuracy of heart disease prediction by applying a Logistic
Regression using a healthcare dataset to determine whether patients have heart illness problems or not.
The dataset was acquired from an ongoing cardiovascular study on people of the town of Framingham,
Massachusetts. The model reached an accuracy prediction of 87 percent. the authors acknowledge the
model could be improved with more data and the use of more ML models.

In [123], the author introduced an accurate classification to examine the breast cancer data with a total of
569 rows and 32 columns, because breast cancer affects one in every 28 women in India. Similarly
employing a heart disease dataset and Lung cancer dataset, this research offered A novel way to
function selection. This method of selection is based on genetic algorithms mixed with the SVM
classification. The classifier results are Lung cancer 81.8182, Diabetes 78.9272. noticed that size, kind,
and source of data used are not indicated.

In [124], the authors, predicted the risk factors that cause heart disease using the K-means clustering
algorithm and analyzed with a visualization tool using a Cleveland heart disease dataset with 76 features
of 303 patients, holds 209 records with 8 attributes such as age, chest pain type, blood pressure, blood
glucose level, ECG in rest, heart rate as well as four types of chest pain. The authors forecast cardiac
diseases by taking into consideration the primary characteristics of four types of chest discomfort solely
and K-means clustering is a common unsupervised ML technique.

In [125], the authors aimed to report on the benefits of various DM methods and proven heart disease
survival prediction models. From the observations, the authors proposed that Logistic Regression and NB
achieved the highest accuracy when performed on a high dimensional dataset on the Cleveland hospital
dataset and DT and RF produce better results on small dimensional datasets. RF delivers more accuracy
than the DT classifier as the algorithm is an optimized learning algorithm. The author mentioned that this
work can be extended to other ML algorithms, the model could be developed in a distributed environment
such as Map-Reduce, Apache Mahout, HBase, etc.

In [126] the authors proposed a single algorithm named hybridization, that combines used techniques
into one single algorithm, The presented Method has three phases, preprocessing phase, classification
phase, and diagnosis phase. They employed the Cleveland database and algorithms NB, SVM, KNN, NN,
J4.8, RF, and GA. NB and SVM always perform better than others, whereas others depend on the specified
features. results attained an accuracy of 89.2 percent. Authors need to enhance accuracy, better accuracy
is the key goal. Notice that the dataset is little, hence the system was not able to train adequately, so the
accuracy of the method was bad.

In [127], the authors presented a study concentrated on the utilization of clinical data for liver disease
prediction and investigate several ways of representing such data through this analysis by utilizing six
algorithms Logistics Regression, KNN, DT, SVM, NB, and RF. The original dataset was taken from the
northeast of Andhra Pradesh, India. includes 583 liver patient’s data whereas 75.64 percent are male and
24.36 percent are female. The analysis result indicated that the Logistics Regression classifier delivers
the most increased order exactness of 75 percent depending on the f1 measure to forecast the liver
Page 19/47
illness and NB gives the least precision of 53 percent. Authors merely studied a few prominent supervised
ML algorithms; more algorithms can be picked to create an increasingly exact model of liver disease
prediction and performance can be steadily improved.

In [128], the authors aimed to predict coronary heart disease (CHD) based on historical medical data
using ML technology. The goal of this study is to use three supervised learning approaches, NB, SVM, and
DT, to find correlations in CHD data that could aid improve prediction rates. The dataset contains a
retrospective sample of males from KEEL, a high-risk heart disease location in the Western Cape of South
Africa. the model utilized NB, SVM, and DT. NB achieved the most accuracy among the three models.
SVM and DT J48 outperformed NB with a specificity rate of 82 percent but showed to have an
inadequate sensitivity rate of less than 50 percent.

In [129], the authors applied data mining and network analysis techniques in hospital admission and
discharge data to analyze the disease or comorbidity footprints of chronic patients. A chronic disease risk
prediction framework was created and evaluated in the Australian healthcare system to predict type 2
diabetes risk. Using a private healthcare funds dataset from Australia that spans six years and three
different predictive algorithms (regression, parameter optimization, and DT). The accuracy of the
prediction ranges from 82 to 87 percent. The hospital admission and discharge summary is the dataset's
source. As a result, it does not provide information about general physician visits or future diagnoses.
2.1.2 DL-based Healthcare Prediction
In [130], the authors proposed a system for predicting the patients with the more common inveterate
diseases with the help of the DL algorithms such as CNN for auto feature extraction and illness prediction
so, they used KNN for distance calculation to locate the exact matching in the dataset and the outcome
of the final prediction of the sickness. A combination of disease symptoms was made for the structure of
the dataset, the living habits of a person, and also the specifies attaches to doctor consultations which
are acceptable in this general disease prediction. In this study, the Indian chronic kidney disease dataset
was utilized that comprises 400 occurrences, 24 characteristics, and 2 classes were restored from the UCI
ML store. At last, a comparative study of the proposed system with other algorithms such as NB, DT, and
logistic regression has been demonstrated in this study. The findings showed that the proposed system
gives an accuracy of 95 percent which is higher than the other two methods. So, the proposed technique
should be applied using more than one dataset.

In [131], the authors developed a DL approach that uses chest radiography images to differentiate
between patients with mild, pneumonia, and COVID-19 infections, providing a valid mechanism for
COVID-19 diagnosis. To increase the intensity of the chest X-ray image and eliminate noise, image-
enhancing techniques were used in the proposed system. Two distinct DL approaches based on a
pertained neural network model (ResNet-50) for COVID-19 identification utilizing Chest X-ray (CXR)
pictures are proposed in this work to minimize overfitting and increase the overall capabilities of the
suggested DL systems. The authors emphasized that tests using a vast and hard dataset encompassing
several COVID-19 cases are necessary to establish the efficacy of the suggested system.
Page 20/47
In [132], the authors presented a Cuckoo search-based deep LSTM classifier for disease prediction. The
deep convLSTM classifier is used in the cuckoo search optimization, which is a nature-inspired method
for accurately predicting disease by transferring information and therefore reducing time consumption.
The PIMA dataset is used to predict the onset of diabetes. The National Institute of Diabetes and
Digestive and Kidney Diseases provided the data. The dataset is made up of independent variables
including insulin level, age, and BMI index, as well as one dependent variable. The new technique was
compared to traditional methods, and the results showed that the proposed method achieved 97.591
percent accuracy, 95.874 percent sensitivity, and 97.094 percent specificity, respectively. authors noticed
more datasets are needed, as well as new approaches to improve the classifier's effectiveness.

In [133], the authors presented a wavelet-based convolutional neural network to handle data limitations in
this time of COVID-19 fast emergence. By investigating the influence of discrete wavelet transform
decomposition up to 4-levels, the model demonstrated the capability of multi-resolution analysis for
detecting COVID-19 Chest X-rays. The wavelet sub-bands are the CNN's inputs at each decomposition
level. COVID-Chest X-ray-12 is a collection of 1,944 chest X-ray pictures divided into 12 groups that were
compiled from two open-source datasets (National Institute Health containing several X-rays of
pneumonia-related diseases where the COVID-19 dataset is collected from Radiology Society North
America). COVID-Neuro wavelet, a suggested model, was trained alongside other well-known ImageNet
pre-trained models on COVID-CXR-12. the authors acknowledge they hope to investigate the effects of
other wavelet functions besides the Haar wavelet.

In [134], the authors developed a CNN framework for COVID-19 identification utilizing computed
tomography images is suggested. The proposed framework employer a public CT dataset of 2482 CT
images from patients of both classifications. the system attained an accuracy of 96.16 percent and recall
of 95.41 percent after training using only 20 percent of the dataset. The authors stated that the use of the
framework should be extended to multimodal medical pictures in the future.

In [135], the authors performed multi-disease prediction for intelligent clinical decision support by
deploying a long short-term memory network and enhancing it with two processes to conduct multi-label
classification based on patients’ clinical visit records. a massive data set of electronic health records
collected from a prominent hospital in southeast china. The suggested LSTM approach outperforms
several standard and DL models in predicting future disease diagnoses, according to model evaluation
results. The F1 score rises from 78.9% and 86.4 percent, respectively, with the state-of-the-art
conventional and DL models, to 88.0 percent with the suggested technique. The authors stated that the
model prediction performance may be enhanced further by including new input variables and that to
reduce computational complexity, the method only uses one data source.

In [136], the authors introduced an approach to creating a supervised ANN structure based on the subnets
(the group of neurons) instead of layers, in the cases of low datasets, this effectively predicted the
disease. The model was evaluated using textual data and compared to Multilayer Perceptron’s (MLPs) as
well as LSTM recurrent neural network models using three small-scale publicly accessible benchmark

Page 21/47
datasets. On the Iris dataset, the experimental findings for classification reached 97 percent accuracy,
compared to 92 percent for RNN (LSTM) with three layers, and the model had a lower error rate, 81, than
RNN (LSTM) and MLP on the diabetic dataset, while RNN (LSTM) has a high error rate of 84. For larger
datasets, however, this method is useless. This model is useless because not implement our model on
large textual and image datasets.

In [137], the authors presented a novel AI and Internet of Things (IoT) convergence-based disease
detection model for a smart healthcare system. Data collection, reprocessing, categorization, and
parameter optimization are all stages of the proposed model. IoT devices, such as wearables and
sensors, collect data, which AI algorithms then use to diagnose diseases. The forest technique is then
used to remove any outliers found in the patient data. Healthcare data was used to assess the
performance of the CSO-LSTM model. During the study, the CSO-LSTM model had a maximum accuracy
of 96.16 percent on heart disease diagnoses and 97.26 percent on diabetes diagnoses. This method
offered a greater prediction accuracy for heart disease and diabetes diagnosis, but there was no feature
selection mechanism, hence it requires extensive computations.

In [138], the authors focused on the coronavirus epidemic, which constitutes a daily threat to global
health. The majority of their research was aimed at detecting disease in people whose Xrays had been
selected as potential COVID-19 candidates. Chest x-rays of people with COVID-19, viral pneumonia, and
healthy people are included in the dataset. The study compared the performance of two DL algorithms,
namely CNN and RNN. DL techniques were used to evaluate a total of 657 chest X-ray images for the
diagnosis of COVID-19. VGG19 is the most successful model, with a 95% accuracy rate. The VGG19
model successfully categorizes COVID-19 patients, healthy individuals, and viral pneumonia cases. The
dataset's most failing approach is InceptionV3. The success percentage can be improved, according to
the authors, by improving data collection. In addition to chest radiography, lung tomography can be used.
The success ratio and performance can be enhanced by creating numerous DL models.

In [139], the authors developed a method based on the RNN algorithm for predicting blood glucose levels
for diabetics a maximum of one hour in the future, which required the patient's glucose level history. The
Ohio T1DM dataset for blood glucose level prediction, which included blood glucose level values for six
people with type 1 diabetes, was used to train and assess the approach. The distribution features were
further honed with the use of studies that revealed the procedure's certainty estimate nature. The authors
point out that they can only evaluate prediction goals with enough glucose level history, thus they can't
anticipate the beginning levels after a gap, which doesn't improve the prediction's quality.

In [140], the authors used an 18-layer residual CNN pre-trained on ImageNet with a different anomaly
detection mechanism for the classification of COVID-19 to construct a new deep anomaly detection
model for speedy, reliable screening. On the X-ray dataset, which contains 100 images from 70 COVID-19
persons and 1431 images from 1008 non-COVID-19 pneumonia subjects, the model obtains a sensitivity
of 90.00 percent specificity of 87.84 percent or sensitivity of 96.00 percent specificity of 70.65 percent.
The authors noted that the model still has certain flaws, such as missing 4% of COVID-19 cases and

Page 22/47
having a 30% false-positive rate. In addition, more clinical data is required to confirm and improve the
model's usefulness.

In [141], the authors developed COVIDX-Net, a novel DL framework that allows radiologists to diagnose
COVID-19 in X-ray images automatically. Seven algorithms (MobileNetV2, ResNetV2, VGG19,
DenseNet201, InceptionV3, Inception, and Xception) were evaluated using a small dataset of 50 photos
(MobileNetV2, ResNetV2, VGG19, DenseNet201, InceptionV3, Inception, and Xception). Each deep neural
network model can classify the patient's status as a negative or positive COVID-19 case based on the
normalized intensities of the X-ray image. The f1-scores for the VGG19 and Dense Convolutional Network
(DenseNet) models were 0.89 and 0.91, respectively. With f1-scores of 0.67, the InceptionV3 model has
the weakest classification performance.

In [142], The authors created a DL approach for delivering 30-minute predictions about future glucose
levels based on a Dilated RNN (DRNN). The performance of the DRNN models was evaluated using data
from two electronic health records datasets: OhioT1DM from clinical trials and the in-silicon dataset from
the UVA-Padova simulator. It outperformed established glucose prediction approaches such as Neural
Networks (NNs), Support Vector Regression (SVR), and autoregressive models (ARX) (ARX). The results
demonstrated that it significantly improved glucose prediction performance, although there are still some
limits, such as the authors' creation of a data-driven model that heavily relies on past EHR. The quality of
the data has a significant impact on the accuracy of the prediction. The number of clinical datasets is
limited, however, often restricted. Because certain data fields are manually entered, they are occasionally
incorrect.

In [143], the authors utilized a deep neural network to discover 15,099 stroke patients, researchers were
able to predict stroke death based on medical history and human behaviors utilizing large-scale electronic
health information. The Korea Centers for Disease Control and Prevention collected data from 2013 to
2016 and found that there are around 150 hospitals in the country, all having more than 100 beds.
Gender, age, type of insurance, mode of admission, necessary brain surgery, area, length of hospital stay,
hospital location, number of hospital beds, stroke kind, and CCI were among the 11 variables in the DL
model. To automatically create features from the data and identify risk factors for stroke, researchers
used a DNN/scaled Principal Component Analysis (PCA). 15,099 people with a history of stroke were
enrolled in the study. The data were divided into a training set (66%) and a testing set (34%), with 30
percent of the samples used for validation in the training set. DNN is used to examine the variables of
interest, while scaled PCA is utilized to improve the DNN's continuous inputs. This study sensitivity,
specificity, and AUC values were respectively 64.32 percent, 85.56 percent, and 83.48 percent.

In [144] the authors proposed a glucose forecasting approach called (GluNet) that used a personalized
DNN for forecasting the probabilistic distribution of short-term measurements having Type 1 diabetes
based on their historical data that involved insulin doses, meal information, glucose measurements, and
various factors. It utilized the newest DL techniques consisting of four components: post-processing,
dilated Convolution Neural Network (CNN), label recovery/ transform, and data pre-processing. authors

Page 23/47
run the models on the subjects from the OhioT1DM datasets. The outcomes revealed significant
enhancements over the previous procedures via a comprehensive comparison concerning the and Root
Mean Square Error (RMSE) having a time lag of 60 mins Prediction Horizons (PH) and RMSE having a
small time lag for the case of prediction horizons in the virtual adult participants. If the PH is properly
matched to the lag between input and output, the user may learn the control of the system more
frequently and it achieves good performance. Additionally, GluNet was validated on two clinical data sets.
It attained an RMSE with a time lag of 60 mins PH and RMSE with a time lag of 30 mins PH. The authors
point out that the model does not consider physiological knowledge, and that they need to test GluNet
with larger prediction horizons and use it to predict overnight hypoglycemia.

In [145], the authors proposed the Short-Term Blood Glucose Prediction Model (VMD-IPSO-LSTM), which
is a short-term strategy for predicting blood glucose (VMD-IPSO-LSTM). Initially, the Intrinsic Modal
Functions (IMF) in various frequency bands were obtained using the Variational Modal Decomposition
(VMD) technique, which deconstructed the blood glucose content. The short and long-term memory
networks then constructed a prediction mechanism for each blood glucose component Intrinsic Modal
Functions (IMF). Because the time window length, learning rate, and neuron count are difficult to set, the
upgraded PSO approach optimized these parameters. The improved LSTM network anticipated each IMF,
and the projected subsequence was superimposed in the final step to arrive at the ultimate prediction
result. The data of 56 participants were chosen as experimental data among 451 diabetic Mellitus
patients. The experiments revealed that it improved prediction accuracy at "30 minutes, 45 minutes, and
60 minutes." The RMSE and MAPE were lower than the "VMD-PSO-LSTM, VMD-LSTM, and LSTM,"
indicating that the suggested model is effective. The longer time it took to anticipate blood glucose levels
and the higher accuracy of the predictions gave patients and doctors more time to improve the
effectiveness of diabetes therapy and manage blood glucose levels. The authors noted that they still
faced challenges, such as an increase in calculation volume and operation time. The time it takes to
estimate glucose levels in the short term will be reduced.

In [146], The authors presented a paradigm for primary COVID-19 detection using a radiology review of
chest radiography or chest X-ray, to reduce diagnosis time and human error. The researchers used a
dataset of chest X-rays from verified COVID-19 patients (408 photographs), confirmed pneumonia
patients (4273 images), and healthy people (1590 images) to perform a three-class image classification
(1590 images). There are 6271 people in total in the dataset. To fulfill this image categorization problem,
the authors plan to use CNN and transfer learning. For all of the folds of data, the model's accuracy
ranged from 93.90 percent to 98.37 percent. Even the lowest level of accuracy, 93.90 percent, is still quite
good. The authors will face a restriction, particularly when it comes to adopting such a model on a large
scale for practical usage.

In [147], the authors proposed DL models for predicting the number of COVID-19 positive cases in Indian
states. The Ministry of Health and Family Welfare dataset contains time-series data for 32 individual
confirmed COVID-19 cases in each of the states (28) and union territories (4) since March 14, 2020. This
dataset was used to conduct an exploratory analysis of the increase in the number of positive cases in
Page 24/47
India. As prediction models, RNN-based LSTMs are used. Deep LSTM, convolutional LSTM, and bi-
directional LSTM models were tested on 32 states/union territories, and the model with the best accuracy
was chosen based on absolute error. Bi-directional LSTM produced the best performance in terms of
prediction errors, while convolutional LSTM produced the worst performance. For all states, daily and
weekly forecasts were calculated, and bi-LSTM produced accurate results (error less than 3%) for short-
term prediction (1–3 days).

In [148], the authors suggested a new type 1 diabetes prediction technique based on CNNs and DL to
improve the robustness and accuracy of type 1 diabetes prediction. It was all about figuring out how to
extract the behavioral pattern. Numerous observations of identical behaviors were used to fill in the gaps
in the data. The suggested model was trained and verified using data from 759 people with type 1
diabetes who visited Sheffield Teaching Hospitals between 2013 and 2015. A subject's type 1 diabetes
test, demographic data (age, gender, years with diabetes), and the final 84 days (12 weeks) of Self-
Monitored Blood Glucose (SMBG) measurements preceding the test formed each item in the training set.
In the presence of insufficient data and certain physiological specificities, prediction accuracy
deteriorates, according to the authors.

In [149], the authors constructed a machine learning technique using the PIDD by NIDDK. PID's
participants are all female and at least 21 years old. PID comprises 768 incidences, with 268 samples
diagnosed as diabetic and 500 samples not diagnosed as diabetic. The eight most important
characteristics that led to diabetes prediction. The accuracy of functional classifiers such as ANN, NB, DT,
and DL is between 90 and 98 percent. On the PIMA dataset, DL had the best results for diabetes onset
among the four, with an accuracy rate of 98.07 percent. The technique uses a variety of classifiers to
accurately predict the disease, but it failed to diagnose it at an early stage.

3. Future Directions
The adoption of ML, as well as DL models, produced a massive impact in all areas and mainly in the
healthcare domain. However, with the availability of numerous reports and scans, human decision-
making remains the only option for diagnosis. This could lead to inaccurate diagnoses due to preference
in human decisions, as a result, many human lives are saved. Therefore, researchers identified numerous
methods for automating diagnosis in healthcare in different specialties. Research has shown that
healthcare professionals are gradually facing the technology for a variety of purposes, such as collecting
patient records or diagnostics. Looking at AI in terms of medical decision-making and data management.
The presented study has drawbacks that can be solved by using more investigations in the future, for
example, there is a possibility that other approaches are present which are not included in this survey.
Next, the use of search keywords like "AI”, "ML", "DL", and "Healthcare" could be common and eliminate the
interesting researches. Furthermore, this survey investigated 40 scientific papers, since the topic of
research is new, the evaluation of more research papers may yield more interesting results.

4. Conclusion
Page 25/47
The use of machine learning, as well as deep learning algorithms for healthcare prediction, has the
potential ability to change the way traditional healthcare services are delivered. In the case of machine
learning and deep learning applications, healthcare data is deemed the most significant component that
contributes to medical-care systems. This paper aims to convey a rich discussion with medical care staff
about how AI can be helpful to them to increase the quality of work. A total of 40 working papers covering
the period from 2019 to 2022 were selected and the methodology for each paper was clarified. studies
have shown that artificial intelligence plays a significant role in diagnosing diseases accurately and helps
to anticipate healthcare and analyze health data by linking hundreds of clinical records and rebuilding a
patient's history using this data. Therefore, there is a need for more studies to improve the links of AI with
data quality management considerations in healthcare.

Abbreviations
AI: Artificial Intelligence; ML: Machine learning; DT: Decision Tree; EHR: Electronic Health Records; RF:
Random Forest; SVM: Support Vector Machine; KNN: K - Nearest Neighbor; NB: Naive Bayes; RL:
Reinforcement learning; NLP: Natural Language Processing; MCTS: Monte Carlo Tree Search; POMDP:
Partially Observable Markov Decision Processes; DL: Deep Learning; DBN: Deep Belief Network; ANNs:
Artificial Neural Networks; CNNs: Convolutional Neural Networks; LSTM: Long Short-Term Memory;
RCNNs: Recurrent Convolution Neural networks; RNNs: Recurrent Neural Networks; RCL: Recurrent
Convolutional Layer; RFs: Receptive Domains; RMLP: Recurrent Multi-Layer Perceptron; PIDD: Pima
Indian Diabetes Database; CHD: Coronary Heart Disease; CXR: chest X-ray; MLPs: Multilayer Perceptron’s;
LSTM: Long Short-Term Memory; IOT: Internet of Things; DRNN: Dilated RNN; NNs: Neural Networks; SVR:
Support Vector Regression; PCA: Principal Component Analysis; PH: Prediction Horizons; RMSE: Root
Mean Square Error; IMF: Intrinsic Modal Functions; VMD: Variational Modal Decomposition; IMF: Intrinsic
Modal Functions; SMBG: Self-Monitored Blood Glucose.

Declarations
Supplementary Information

Not applicable.

Acknowledgments

Not applicable.

Authors’ contributions

All authors have participated equally in this work.

Funding

Not applicable.
Page 26/47
Availability of data and materials

The corresponding author can provide the material used and data analyzed on request.

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests. All authors approved the final manuscript.

References
1. Hema Latha M, Ramakrishna A, Sudarsha Chakravarthi Reddy B, Venkateswarlu C, Yamini
Saraswathi S. Disease Prediction by Stacking Algorithms Over Big Data from Healthcare
Communities. InIntelligent Manufacturing and Energy Sustainability 2022 (pp. 355–363). Springer,
Singapore.
2. Elmahdy HN. Medical Diagnosis Enhancements through Artificial Intelligence.
3. Van Calster B, Wynants L, Timmerman D, Steyerberg EW, Collins GS. Predictive analytics in health
care: how can we know it works?. Journal of the American Medical Informatics Association. 2019
Dec;26(12):1651–4.
4. Sahoo PK, Mohapatra SK, Wu SL. SLA based healthcare big data analysis and computing in cloud
network. Journal of Parallel and Distributed Computing. 2018 Sep 1;119:121 – 35.
5. Thanigaivasan V, Narayanan SJ, Iyengar SN, Ch N. Analysis of parallel SVM based classification
technique on healthcare using big data management in cloud storage. Recent Patents on Computer
Science. 2018 Aug 1;11(3):169 – 78.
6. Wang Y, Kung L, Wang WY, Cegielski CG. An integrated big data analytics-enabled transformation
model: Application to health care. Information & Management. 2018 Jan 1;55(1):64–79.
7. Omran, H. Primary health care and health care administration. In: Recent Patents on University of
Basrah11.3, 2016. DOI:10.13140/RG.2.2.33481.34406.
8. Xiong X, Cao X, Luo L. The ecology of medical care in Shanghai. BMC Health Services Research.
2021 Dec;21(1):1–9.
9. Burazeri G, Kragelj LZ. Health: Systems–Lifestyle–Policies. (Volume I)Edition: 2ndChapter: The role
and organization of health care systems.
10. Marzorati C, Pravettoni G. Value as the key concept in the health care system: how it has influenced
medical practice and clinical decision-making processes. Journal of multidisciplinary healthcare.
Page 27/47
 2017;10:101.
11. Qayyum A, Qadir J, Bilal M, Al-Fuqaha A. Secure and robust machine learning for healthcare: A
survey. IEEE Reviews in Biomedical Engineering. 2020 Jul 31;14:156–80.
12. El Seddawy AB, Moawad R, Hana MA. Applying Data Mining Techniques in CRM.
13. Malik M, Khatana R, Kaushik A. Machine Learning With Health Care: A perspective. InJournal of
Physics: Conference Series 2021 Oct 1 (Vol. 2040, No. 1, p. 012022). IOP Publishing.
14. Mirbabaie M, Stieglitz S, Frick NR. Artificial intelligence in disease diagnostics: A critical review and
classification on the current state of research guiding future direction. Health and Technology. 2021
Jul;11(4):693–731.
15. Singh G, Al’Aref SJ, Van Assen M, Kim TS, van Rosendael A, Kolli KK, Dwivedi A, Maliakal G, Pandey
M, Wang J, Do V. Machine learning in cardiac CT: basic concepts and contemporary data. Journal of
Cardiovascular Computed Tomography. 2018 May 1;12(3):192–201.
16. Kim KJ, Tagkopoulos I. Application of machine learning in rheumatic disease research. The Korean
journal of internal medicine. 2019 Jul;34(4):708.
17. Liu B. Supervised learning. InWeb data mining 2011 (pp. 63–132). Springer, Berlin, Heidelberg.
18. Haykin S, Lippmann R. Neural networks, a comprehensive foundation. International journal of neural
systems. 1994;5(4):363–4.
19. Monica, G.. A Comparative Study on Supervised Machine Learning Algorithm. International Journal
for Research in Applied Science & Engineering Technology (IJRASET) 2022.
20. Ray S. A quick review of machine learning algorithms. In2019 International conference on machine
learning, big data, cloud and parallel computing (COMITCon) 2019 Feb 14 (pp. 35–39). IEEE.
21. Srivastava A, Saini S, Gupta D. Comparison of various machine learning techniques and its uses in
different fields. In2019 3rd International conference on electronics, communication and aerospace
technology (ICECA) 2019 Jun 12 (pp. 81–86). IEEE.
22. Park HA. An introduction to logistic regression: from basic concepts to interpretation with particular
attention to nursing domain. Journal of Korean Academy of Nursing. 2013 Apr 1;43(2):154 – 64.
23. Obulesu O, Mahendra M, ThrilokReddy M. Machine learning techniques and tools: A survey. In2018
International Conference on Inventive Research in Computing Applications (ICIRCA) 2018 Jul 11
(pp. 605–611). IEEE.
24. Dhall D, Kaur R, Juneja M. Machine learning: a review of the algorithms and its applications.
Proceedings of ICRIC 2019. 2020:47–63.
25. Yang FJ. An extended idea about decision trees. In2019 International Conference on Computational
Science and Computational Intelligence (CSCI) 2019 Dec 5 (pp. 349–354). IEEE.
26. Eesa AS, Orman Z, Brifcani AM. A novel feature-selection approach based on the cuttlefish
optimization algorithm for intrusion detection systems. Expert systems with applications. 2015 Apr
1;42(5):2670-9.

Page 28/47
27. Shamim A, Hussain H, Shaikh MU. A framework for generation of rules from decision tree and
decision table. In2010 International Conference on Information and Emerging Technologies 2010
Jun 14 (pp. 1–6). IEEE.
28. Eesa AS, Abdulazeez AM, Orman Z. A DIDS Based on The Combination of Cuttlefish Algorithm and
Decision Tree. Science Journal of University of Zakho. 2017 Dec 30;5(4):313–8.
29. Bakyarani S, Srimathi H, & Bagavandas M. a survey of machine learning algorithms in health care.
international journal of scientific and technical research, volume 8, issue 11, November 2019.
30. Resende PA, Drummond AC. A survey of random forest based methods for intrusion detection
systems. ACM Computing Surveys (CSUR). 2018 May 23;51(3):1–36.
31. Breiman L. Random forests. Machine learning. 2001 Oct;45(1):5–32.
32. Ho TK. The random subspace method for constructing decision forests. IEEE transactions on pattern
analysis and machine intelligence. 1998 Aug;20(8):832–44.
33. Hofmann M, Klinkenberg R, editors. RapidMiner: Data mining use cases and business analytics
applications. CRC Press; 2016 Apr 19.
34. Chow CK, Liu C. Approximating discrete probability distributions with dependence trees. IEEE
transactions on Information Theory. 1968 May;14(3):462–7.
35. Burges CJ. A tutorial on support vector machines for pattern recognition. Data mining and
knowledge discovery. 1998 Jun;2(2):121–67.
36. Han J, Kamber M, Mining D. Data Mining Concepts and Techniques, Elevier, 2011.
37. Cortes C, Vapnik V. Support-vector networks. Machine learning. 1995 Sep;20(3):273–97.
38. Aldahiri A, Alrashed B, Hussain W. Trends in using IoT with machine learning in health prediction
system. Forecasting. 2021 Mar 7;3(1):181–206.
39. Sarker IH. Machine learning: Algorithms, real-world applications and research directions. SN
Computer Science. 2021 May;2(3):1–21.
40. Ting KM, Zheng Z. Improving the performance of boosting for naive Bayesian classification.
InPacific-Asia Conference on Knowledge Discovery and Data Mining 1999 Apr 26 (pp. 296–305).
Springer, Berlin, Heidelberg.
41. Kaur R, Juneja M. A survey of different imaging modalities for renal cancer. Indian J Sci Technol.
2016 Nov;9(44):1–6.
42. Shailaja K, Seetharamulu B, Jabbar MA. Machine learning in healthcare: A review. In2018 Second
international conference on electronics, communication and aerospace technology (ICECA) 2018
Mar 29 (pp. 910–914). IEEE.
43. Mahesh B. Machine learning algorithms-a review. International Journal of Science and Research
(IJSR).[Internet]. 2020 Oct;9:381–6.
44. Greene D, Cunningham P, Mayer R. Unsupervised learning and clustering. InMachine learning
techniques for multimedia 2008 (pp. 51–90). Springer, Berlin, Heidelberg.
45. Jain AK, Dubes RC. Algorithms for clustering data. Prentice-Hall, Inc.; 1988 Jul 1.
Page 29/47
46. Kodinariya TM, Makwana PR. Review on determining number of Cluster in K-Means Clustering.
International Journal. 2013 Nov;1(6):90–5.
47. Shlens J. A tutorial on principal component analysis. arXiv preprint arXiv:1404.1100. 2014 Apr 3.
48. Mishra SP, Sarkar U, Taraphder S, Datta S, Swain D, Saikhom R, Panda S, Laishram M. Multivariate
statistical data analysis-principal component analysis (PCA). International Journal of Livestock
Research. 2017 May;7(5):60–78.
49. Kamani MM, Haddadpour F, Forsati R, Mahdavi M. Efficient fair principal component analysis.
Machine Learning. 2022 Jan 6:1–32.
50. Dey, A. machine learning algorithms: a review. International Journal of Computer Science and
Information Technologies, 2016.
51. Agrawal R, Imieliński T, Swami A. Mining association rules between sets of items in large databases.
InProceedings of the 1993 ACM SIGMOD international conference on Management of data 1993 Jun
1 (pp. 207–216).
52. Agrawal R, Srikant R. Fast algorithms for mining association rules. InProc. 20th int. conf. very large
data bases, VLDB 1994 Sep 12 (Vol. 1215, pp. 487–499).
53. Singh J, Ram H, Sodhi DJ. Improving efficiency of apriori algorithm using transaction reduction.
International Journal of Scientific and Research Publications. 2013 Jan;3(1):1–4.
54. Al-Maolegi M, Arkok B. An improved Apriori algorithm for association rules. arXiv preprint
arXiv:1403.3948. 2014 Mar 16.
55. Abaya SA. Association rule mining based on Apriori algorithm in minimizing candidate generation.
International Journal of Scientific & Engineering Research. 2012 Jul;3(7):1–4.
56. Coronato A, Naeem M, De Pietro G, Paragliola G. Reinforcement learning for intelligent healthcare
applications: A survey. Artificial Intelligence in Medicine. 2020 Sep 1;109:101964.
57. Watkins CJ. Learning from delayed rewards.
58. Chapman D, Kaelbling LP. Input Generalization in Delayed Reinforcement Learning: An Algorithm and
Performance Comparisons. InIjcai 1991 Aug 24 (Vol. 91, pp. 726–731).
59. Watkins CJ, Dayan P. Q-learning. Machine learning. 1992 May;8(3):279 – 92.
60. Jang B, Kim M, Harerimana G, Kim JW. Q-learning algorithms: A comprehensive classification and
applications. IEEE access. 2019 Sep 13;7:133653–67.
61. Achille A, Soatto S. Information dropout: Learning optimal representations through noisy
computation. IEEE transactions on pattern analysis and machine intelligence. 2018 Jan
10;40(12):2897 – 905.
62. Williams G, Wagener N, Goldfain B, Drews P, Rehg JM, Boots B, Theodorou EA. Information theoretic
MPC for model-based reinforcement learning. In2017 IEEE International Conference on Robotics and
Automation (ICRA) 2017 May 29 (pp. 1714–1721). IEEE.
63. Wilkes J, Gallistel CR. Information theory, memory, prediction, and timing in associative learning.

Page 30/47
64. Jang B, Kim M, Harerimana G, Kim JW. Q-learning algorithms: A comprehensive classification and
applications. IEEE access. 2019 Sep 13;7:133653–67.
65. Ning Y, Jia J, Wu Z, Li R, An Y, Wang Y, Meng H. Multi-task deep learning for user intention
understanding in speech interaction systems. InThirty-First AAAI Conference on Artificial Intelligence
2017 Feb 10.
66. Shi X, Gao Z, Lausen L, Wang H, Yeung DY, Wong WK, Woo WC. Deep learning for precipitation
nowcasting: A benchmark and a new model. Advances in neural information processing systems.
2017;30.
67. Juang CF, Lu CM. Ant colony optimization incorporated with fuzzy Q-learning for reinforcement fuzzy
control. IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans. 2009
Feb 27;39(3):597–608.
68. Świechowski M, Godlewski K, Sawicki B, Mańdziuk J. Monte carlo tree search: A review of recent
modifications and applications. Artificial Intelligence Review. 2022 Jul 19:1–66.
69. Lizotte DJ, Laber EB. Multi-objective Markov decision processes for data-driven decision support.
The Journal of Machine Learning Research. 2016 Jan 1;17(1):7378 – 405.
70. Silver D, Huang A, Maddison CJ, Guez A, Sifre L, Van Den Driessche G, Schrittwieser J, Antonoglou I,
Panneershelvam V, Lanctot M, Dieleman S. Mastering the game of Go with deep neural networks and
tree search. nature. 2016 Jan;529(7587):484–9.
71. Baier H, Drake PD. The power of forgetting: Improving the last-good-reply policy in Monte Carlo Go.
IEEE Transactions on Computational Intelligence and AI in Games. 2010 Dec 20;2(4):303–9.
72. Browne CB, Powley E, Whitehouse D, Lucas SM, Cowling PI, Rohlfshagen P, Tavener S, Perez D,
Samothrakis S, Colton S. A survey of monte carlo tree search methods. IEEE Transactions on
Computational Intelligence and AI in games. 2012 Feb 3;4(1):1–43.
73. Ling ZH, Kang SY, Zen H, Senior A, Schuster M, Qian XJ, Meng HM, Deng L. Deep learning for
acoustic modeling in parametric speech generation: A systematic review of existing techniques and
future trends. IEEE Signal Processing Magazine. 2015 Apr 2;32(3):35–52.
74. Schmidhuber J. Deep learning in neural networks: An overview. Neural networks. 2015 Jan 1;61:85–
117.
75. Yu D, Deng L. Deep learning and its applications to signal and information processing [exploratory
dsp]. IEEE Signal Processing Magazine. 2010 Dec 17;28(1):145–54.
76. Hinton GE, Osindero S, Teh YW. A fast learning algorithm for deep belief nets. Neural computation.
2006 Jul 1;18(7):1527–54.
77. Goyal P, Pandey S, Jain K. Introduction to natural language processing and deep learning. InDeep
Learning for Natural Language Processing 2018 (pp. 1–74). Apress, Berkeley, CA.
78. Mathew A, Amudha P, Sivakumari S. Deep learning techniques: an overview. InInternational
conference on advanced machine learning technologies and applications 2020 Feb 13 (pp. 599–
608). Springer, Singapore.

Page 31/47
79. Bengio A, Yoshua G. ian, Courville,“Deep learning,”. Nature. 2015;29(7553):1–73.
80. Gomes L. Machine-learning maestro michael jordan on the delusions of big data and other huge
engineering efforts. IEEE spectrum. 2014 Oct 20;20.
81. Huang G, Liu Z, Van Der Maaten L, Weinberger KQ. Densely connected convolutional networks.
InProceedings of the IEEE conference on computer vision and pattern recognition 2017 (pp. 4700–
4708).
82. Yap MH, Pons G, Marti J, Ganau S, Sentis M, Zwiggelaar R, Davison AK, Marti R. Automated breast
ultrasound lesions detection using convolutional neural networks. IEEE journal of biomedical and
health informatics. 2017 Aug 7;22(4):1218-26.
83. Goodfellow I, Bengio Y, Courville A. Deep learning. MIT press; 2016 Nov 10.
84. Alom MZ, Taha TM, Yakopcic C, Westberg S, Sidike P, Nasrin MS, Hasan M, Van Essen BC, Awwal AA,
Asari VK. A state-of-the-art survey on deep learning theory and architectures. Electronics. 2019
Mar;8(3):292.
85. Apaydin H, Feizi H, Sattari MT, Colak MS, Shamshirband S, Chau KW. Comparative analysis of
recurrent neural network architectures for reservoir inflow forecasting. Water. 2020 May
24;12(5):1500.
86. Ganatra N, Patel A. A comprehensive study of deep learning architectures, applications and tools.
International Journal of Computer Sciences and Engineering. 2018;6(12):701–5.
87. Graves A, Mohamed AR, Hinton G. Speech recognition with deep recurrent neural networks. In2013
IEEE international conference on acoustics, speech and signal processing 2013 May 26 (pp. 6645–
6649). Ieee.
88. Bengio Y, Simard P, Frasconi P. Learning long-term dependencies with gradient descent is difficult.
IEEE transactions on neural networks. 1994 Mar;5(2):157–66.
89. Min S, Lee B, Yoon S. Deep learning in bioinformatics. Briefings in bioinformatics. 2017 Sep
1;18(5):851–69.
90. Glorot X, Bengio Y. Understanding the difficulty of training deep feedforward neural networks.
InProceedings of the thirteenth international conference on artificial intelligence and statistics 2010
Mar 31 (pp. 249–256). JMLR Workshop and Conference Proceedings.
91. Alzubaidi L, Zhang J, Humaidi AJ, Al-Dujaili A, Duan Y, Al-Shamma O, Santamaría J, Fadhel MA, Al-
Amidie M, Farhan L. Review of deep learning: Concepts, CNN architectures, challenges, applications,
future directions. Journal of big Data. 2021 Dec;8(1):1–74.
92. Hochreiter S, Schmidhuber J. Long short-term memory. Neural computation. 1997 Nov 15;9(8):1735–
80.
93. Smagulova K, James AP. A survey on LSTM memristive neural network architectures and
applications. The European Physical Journal Special Topics. 2019 Oct;228(10):2313–24.
94. Setyanto A, Laksito A, Alarfaj F, Alreshoodi M, Oyong I, Hayaty M, Alomair A, Almusallam N,
Kurniasari L. Arabic Language Opinion Mining Based on Long Short-Term Memory (LSTM). Applied

Page 32/47
 Sciences. 2022 Jan;12(9):4140.
95. Lindemann B, Müller T, Vietz H, Jazdi N, Weyrich M. A survey on long short-term memory networks
for time series prediction. Procedia CIRP. 2021 Jan 1;99:650–5.
96. Cui Z, Ke R, Pu Z, Wang Y. Deep bidirectional and unidirectional LSTM recurrent neural network for
network-wide traffic speed prediction. arXiv preprint arXiv:1801.02143. 2018 Jan 7.
97. Villegas R, Yang J, Zou Y, Sohn S, Lin X, Lee H. Learning to generate long-term future via hierarchical
prediction. Ininternational conference on machine learning 2017 Jul 17 (pp. 3560–3569). PMLR.
98. Gensler A, Henze J, Sick B, Raabe N. Deep Learning for solar power forecasting—An approach using
AutoEncoder and LSTM Neural Networks. In2016 IEEE international conference on systems, man,
and cybernetics (SMC) 2016 Oct 9 (pp. 002858–002865). IEEE.
99. Lindemann B, Fesenmayr F, Jazdi N, Weyrich M. Anomaly detection in discrete manufacturing using
self-learning approaches. Procedia CIRP. 2019 Jan 1;79:313–8.
100. Kalchbrenner N, Danihelka I, Graves A. Grid long short-term memory. arXiv preprint arXiv:1507.01526.
2015 Jul 6.
101. Cheng B, Xu X, Zeng Y, Ren J, Jung S. Pedestrian trajectory prediction via the Social-Grid LSTM
model. The Journal of Engineering. 2018 Nov;2018(16):1468–74.
102. Veličković P, Karazija L, Lane ND, Bhattacharya S, Liberis E, Liò P, Chieh A, Bellahsen O, Vegreville M.
Cross-modal recurrent models for weight objective prediction from multimodal time-series data.
InProceedings of the 12th EAI International Conference on Pervasive Computing Technologies for
Healthcare 2018 May 21 (pp. 178–186).
103. Wang J, Hu X. Convolutional neural networks with gated recurrent connections. IEEE Transactions on
Pattern Analysis and Machine Intelligence. 2021 Jan 26.
104. Liang M, Hu X. Recurrent convolutional neural network for object recognition. InProceedings of the
IEEE conference on computer vision and pattern recognition 2015 (pp. 3367–3375).
105. Liang M, Hu X, Zhang B. Convolutional neural networks with intra-layer recurrent connections for
scene labeling. Advances in neural information processing systems. 2015;28.
106. Fernandez B, Parlos AG, Tsai WK. Nonlinear dynamic system identification using artificial neural
networks (ANNs). In1990 IJCNN international joint conference on neural networks 1990 Jun 17
(pp. 133–141). IEEE.
107. Puskorius GV, Feldkamp LA. Neurocontrol of nonlinear dynamical systems with Kalman filter trained
recurrent networks. IEEE Transactions on neural networks. 1994 Mar;5(2):279–97.
108. Rumelhart DE, McClelland JL, PDP Research Group. Parallel Distributed Processing: Explorations in
the Microstructure of Cognition two voll.
109. Krishnamoorthi R, Joshi S, Almarzouki HZ, Shukla PK, Rizwan A, Kalpana C, Tiwari B. A novel
diabetes healthcare disease prediction framework using machine learning techniques. Journal of
Healthcare Engineering. 2022 Jan 11;2022.

Page 33/47
110. Edeh MO, Khalaf OI, Tavera CA, Tayeb S, Ghouali S, Abdulsahib GM, Richard-Nnabu NE, Louni A. A
classification algorithm-based hybrid diabetes prediction model. Frontiers in Public Health. 2022;10.
111. Iwendi C, Huescas CG, Chakraborty C, Mohan S. COVID-19 health analysis and prediction using
machine learning algorithms for Mexico and Brazil patients. Journal of Experimental & Theoretical
Artificial Intelligence. 2022 Apr 7:1–21.
112. Lu, H., Uddin, S., Hajati, F., Moni, M. A., & Khushi, M. (2022). A patient network-based machine
learning model for disease prediction: The case of type 2 diabetes mellitus. Applied Intelligence,
52(3), 2411–2422.
113. Chugh M, Johari R, Goel A. MATHS: Machine Learning Techniques in Healthcare System.
InInternational Conference on Innovative Computing and Communications 2022 (pp. 693–702).
Springer, Singapore.
114. Deberneh HM, Kim I. Prediction of Type 2 diabetes based on machine learning algorithm.
International journal of environmental research and public health. 2021 Mar 23;18(6):3317.
115. Gupta S, Verma HK, Bhardwaj D. Classification of diabetes using Naive Bayes and support vector
machine as a technique. InOperations Management and Systems Engineering 2021 (pp. 365–376).
Springer, Singapore.
116. Islam MT, Rafa SR, Kibria MG. Early prediction of heart disease using PCA and hybrid genetic
algorithm with k-means. In2020 23rd International Conference on Computer and Information
Technology (ICCIT) 2020 Dec 19 (pp. 1–6). IEEE.
117. Qawqzeh YK, Bajahzar AS, Jemmali M, Otoom MM, Thaljaoui A. Classification of diabetes using
photoplethysmogram (PPG) waveform analysis: Logistic regression modeling. BioMed Research
International. 2020 Aug 11;2020.
118. Grampurohit S, Sagarnal C. Disease prediction using machine learning algorithms. In2020
International Conference for Emerging Technology (INCET) 2020 Jun 5 (pp. 1–7). IEEE.
119. Moturi S, Srikanth Vemuru DS. Classification model for prediction of heart disease using correlation
coefficient technique. International Journal. 2020 Mar;9(2).
120. Barik S, Mohanty S, Rout D, Mohanty S, Patra AK, Mishra AK. Heart disease prediction using machine
learning techniques. InAdvances in Electrical Control and Signal Systems 2020 (pp. 879–888).
Springer, Singapore.
121. Princy RJ, Parthasarathy S, Jose PS, Lakshminarayanan AR, Jeganathan S. Prediction of cardiac
disease using supervised machine learning algorithms. In2020 4th international conference on
intelligent computing and control systems (ICICCS) 2020 May 13 (pp. 570–575). IEEE.
122. Saw M, Saxena T, Kaithwas S, Yadav R, Lal N. Estimation of prediction for getting heart disease
using logistic regression model of machine learning. In2020 International Conference on Computer
Communication and Informatics (ICCCI) 2020 Jan 22 (pp. 1–6). IEEE.
123. Soni VD. Chronic disease detection model using machine learning techniques. International Journal
of Scientific & Technology Research. 2020 Sep;9(9):262–6.

Page 34/47
124. Indrakumari R, Poongodi T, Jena SR. Heart disease prediction using exploratory data analysis.
Procedia Computer Science. 2020 Jan 1;173:130–9.
125. Wu CS, Badshah M, Bhagwat V. Heart disease prediction using data mining techniques.
InProceedings of the 2019 2nd international conference on data science and information technology
2019 Jul 19 (pp. 7–11).
126. Tarawneh M, Embarak O. Hybrid approach for heart disease prediction using data mining techniques.
InInternational Conference on Emerging Internetworking, Data & Web Technologies 2019 Feb 26
(pp. 447–454). Springer, Cham.
127. Rahman AS, Shamrat FJ, Tasnim Z, Roy J, Hossain SA. A comparative study on liver disease
prediction using supervised machine learning algorithms. International Journal of Scientific &
Technology Research. 2019 Nov;8(11):419–22.
128. Gonsalves AH, Thabtah F, Mohammad RM, Singh G. Prediction of coronary heart disease using
machine learning: an experimental analysis. InProceedings of the 2019 3rd International Conference
on Deep Learning Technologies 2019 Jul 5 (pp. 51–56).
129. Khan A, Uddin S, Srinivasan U. Chronic disease prediction using administrative data and graph
theory: The case of type 2 diabetes. Expert Systems with Applications. 2019 Dec 1;136:230 – 41.
130. Alanazi R. Identification and prediction of chronic diseases using machine learning approach.
Journal of Healthcare Engineering. 2022 Feb 25;2022.
131. Gouda W, Almurafeh M, Humayun M, Jhanjhi NZ. Detection of COVID-19 Based on Chest X-rays
Using Deep Learning. InHealthcare 2022 Feb 10 (Vol. 10, No. 2, p. 343). MDPI.
132. Kumar A, Satyanarayana Reddy SS, Mahommad GB, Khan B, Sharma R. Smart Healthcare: Disease
Prediction Using the Cuckoo-Enabled Deep Classifier in IoT Framework. Scientific Programming.
2022 May 6;2022.
133. Li JP, Nneji GU, James EC, Chikwendu IA, Ejiyi CJ, Oluwasanmi A, Mgbejime GT. The capability of
multi resolution analysis: A case study of COVID-19 diagnosis. In2021 4th International Conference
on Pattern Recognition and Artificial Intelligence (PRAI) 2021 Aug 20 (pp. 236–242). IEEE.
134. Al Rahhal MM, Bazi Y, Jomaa RM, Zuair M, Al Ajlan N. Deep learning approach for COVID-19
detection in computed tomography images. Cmc-Computers Materials & Continua. 2021:2093–110.
135. Men L, Ilk N, Tang X, Liu Y. Multi-disease prediction using LSTM recurrent neural networks. Expert
Systems with Applications. 2021 Sep 1;177:114905.
136. Ahmad U, Song H, Bilal A, Mahmood S, Alazab M, Jolfaei A, Ullah A, Saeed U. A novel deep learning
model to secure internet of things in healthcare. InMachine intelligence and big data analytics for
cybersecurity applications 2021 (pp. 341–353). Springer, Cham.
137. Mansour RF, El Amraoui A, Nouaouri I, Díaz VG, Gupta D, Kumar S. Artificial intelligence and internet
of things enabled disease diagnosis model for smart healthcare systems. IEEE Access. 2021 Mar
17;9:45137–46.
138. Sevi M, Aydin İ. COVID-19 detection using deep learning methods. In2020 International conference on
data analytics for business and industry: way towards a sustainable economy (ICDABI) 2020 Oct 26
Page 35/47
 (pp. 1–6). IEEE.
139. Martinsson J, Schliep A, Eliasson B, Mogren O. Blood glucose prediction with variance estimation
using recurrent neural networks. Journal of Healthcare Informatics Research. 2020 Mar;4(1):1–8.
140. Zhang J, Xie Y, Li Y, Shen C, Xia Y. Covid-19 screening on chest x-ray images using deep learning
based anomaly detection. arXiv preprint arXiv:2003.12338. 2020 Mar 27;27.
141. Hemdan EE, Shouman MA, Karar ME. Covidx-net: A framework of deep learning classifiers to
diagnose covid-19 in x-ray images. arXiv preprint arXiv:2003.11055. 2020 Mar 24.
142. Zhu T, Li K, Chen J, Herrero P, Georgiou P. Dilated recurrent neural networks for glucose forecasting in
type 1 diabetes. Journal of Healthcare Informatics Research. 2020 Sep;4(3):308–24.
143. Cheon S, Kim J, Lim J. The use of deep learning to predict stroke patient mortality. International
journal of environmental research and public health. 2019 Jun;16(11):1876.
144. Li K, Liu C, Zhu T, Herrero P, Georgiou P. GluNet: A deep learning framework for accurate glucose
forecasting. IEEE journal of biomedical and health informatics. 2020 Jul 29;24(2):414–23.
145. Wang W, Tong M, Yu M. Blood glucose prediction with VMD and LSTM optimized by improved
particle swarm optimization. IEEE Access. 2020 Dec 4;8:217908–16.
146. Rashid N, Hossain MA, Ali M, Sukanya MI, Mahmud T, Fattah SA. Transfer Learning Based Method
for COVID-19 Detection From Chest X-ray Images. In2020 IEEE REGION 10 CONFERENCE (TENCON)
2020 Nov 16 (pp. 585–590). IEEE.
147. Arora P, Kumar H, Panigrahi BK. Prediction and analysis of COVID-19 positive cases using deep
learning models: A descriptive case study of India. Chaos, Solitons & Fractals. 2020 Oct
1;139:110017.
148. Zaitcev A, Eissa MR, Hui Z, Good T, Elliott J, Benaissa M. A deep neural network application for
improved prediction of $\text {HbA} _ {\text {1c}} $ in type 1 diabetes. IEEE journal of biomedical and
health informatics. 2020 Jan 17;24(10):2932–41.
149. Naz H, Ahuja S. Deep learning approach for diabetes prediction using PIMA Indian dataset. Journal
of Diabetes & Metabolic Disorders. 2020 Jun;19(1):391–403.

Figures

Page 36/47
Figure 1

Illustration of heterogeneous sources contributing to healthcare data [11].

Page 37/47
Figure 2

AI, ML, and DL.

Page 38/47
Figure 3

Different types of Machine Learning algorithms [16].

Page 39/47
Figure 4

Linear Regression model [21].

Page 40/47
Figure 5

Example of a DT.

Figure 6
Page 41/47
Random Forest architecture [21].

Figure 7

Support Vector Machine [19].

Figure 8

Page 42/47
K-Nearest Neighbor [19].

Figure 9

Workflow of unsupervised learning [24].

Figure 10

Visualization of data before and after applying PCA [50].

Page 43/47
Figure 11

The basic MCTS process [71].

Page 44/47
Figure 12

The performance of deep learning concerning the amount of data [78].

Figure 13

The architecture of CNN [84].

Page 45/47
Figure 14

LSTM unit. [94].

Figure 15

(left) Bidirectional LSTM and (right) filter mechanism for processing incomplete data [96].

Page 46/47
Figure 16

Illustration of the architectures of CNN, RMLP, and RCNN [97].

Page 47/47