INTRODUCTION

Cardiovascular disease describes various conditions that can affect the human heart. Heart
disease is most complex human diseases across the globe. According to reports from the
World Health Organization (WHO), cardiovascular disease kills 17.9 million people per
year globally. Claims that in heart disease, the heart pumps insufficient amounts of blood
to other body organs, which affects their functionalities. Accordingly, some of the
activities that increase the likelihood of developing heart disease are obesity, high levels of
cholesterol, and high blood pressure, among others. In addition, age, genetics, and past
events also influence the likelihood of developing heart disease. As described by the
American Heart Association, individuals suffering from heart disease show various signs
and symptoms. These people experience challenges in their sleep, an irregular heartbeat
(heart rate decrease or increase), rapid weight loss, and swollen legs. However, these signs
and symptoms are common for different diseases, especially in elderly people. Therefore,
it is difficult to get the actual diagnosis, which may lead to increased mortality soon.
A correct diagnosis of heart disease is critical to reducing mortality. Prediction helps
Physicians often use the angiography approach to diagnose heart disease. However, this
diagnostic approach is time-consuming and cost-ineffective, especially in developing
countries where healthcare providers, diagnostic technologies, and other resources are
limited. In recent years, the health industry has incorporated modern technology to offer
better services to patients. With the modern technological advances in the health industry,
patients' data can easily be accessed through several available open sources. Using this
data, research can be carried out so that different modern technologies can be used to
correctly diagnose patients and detect heart disease before the condition worsens.
Artificial intelligence and machine learning are critical in the prediction and detection of
heart diseases. Different models of deep learning and machine learning can be used to
diagnose cardiovascular disease and predict outcomes. Researchers use different machine
learning techniques to conduct comprehensive genomic data analysis within a short time
and with high accuracy.
Traditional diagnostic methods for heart disease typically involve invasive techniques that
rely on a comprehensive evaluation of the patient's medical history, a physical
examination, and a thorough analysis of the patient's symptoms by medical professionals.
Despite the significant advancements in medical science and technology, these traditional
methods still have inherent limitations, including inaccuracies and delays in diagnosis
results, which can be attributed to human error. Furthermore, the use of these traditional
diagnostic methods often requires a significant number of financial resources, as well as
advanced computational and technical expertise, and can also be time-consuming, leading
to additional stress and anxiety for patients.
This report analyse a dataset containing information about five different heart diseases.
The data set is representative of a single large data set on cardiovascular disease, thanks to
the inclusion of twelve standard features. Researchers can use methods like machine
learning on the dataset to learn more about the trend, identify the most at-risk populations,
and discover other insights. This will help the health ministry better provide care for
patients suffering from heart disease by predicting the earliest stages of the disease. The
dataset we are using to build a heart prediction model contains 303 rows and 14 columns
with all health-related data. We will use the same dataset to check the accuracy of the
prediction algorithm, which was developed using logistic regression.
Heart disease prediction allows healthcare providers to make informed decisions about the
health of a patient. Using machine learning helps to understand and reduce the symptoms
of cardiovascular diseases. Heart disease can be predicted using the multiple regression
model, demonstrating the validity of multiple logical regression. The work is done on a
data set of 1026 instances with 14 different attributes. 70% of the data is used for training
purposes, while the remaining 30% is used for validation.
When working with machine learning, dealing with the high dimensionality of data is a
common challenge. With datasets that contain huge amounts of data, it can be difficult to
even visualize the data in three dimensions, which is referred to as the curse of
dimensionality. The processing of such large datasets can require a huge amount of
memory and can lead to issues such as overfitting. However, one approach to addressing
this issue is to use weighting features, which can decrease the redundancy in the dataset
and reduce the processing time required for execution. To further tackle the dimensionality
of the dataset, there are various techniques for feature engineering and feature selection
that can be utilized to remove data that may not be as important in the overall dataset.
Data mining techniques have been widely used in healthcare to predict and diagnose
chronic diseases based on previous health records. Different algorithms, for example
Naive Bayes, Classification Tree, ANN, SVM, and Logistic Regression, can be used in
predicting cardiovascular diseases. Compared to other algorithms, Logistic Regression has
the highest level of precision.
Machine Learning (ML)
Machine learning is widely used in almost all fields in the world, including the healthcare
sector. Machine learning is an application of artificial intelligence (AI) that provides
systems with the ability to automatically learn and improve from experience without being
explicitly programmed. Further, machine learning at its most basic is the practice of using
algorithms to parse data learn from it and then decide or make predictions about
something in the world. There are two major categories of problems often solved by
machine learning, i.e., regression and classification. Mainly, regression algorithms are
used for numeric data and classification problems include binary and multi-category
problems. Machine learning algorithms are further divided into two categories, such as
supervised learning and unsupervised learning. Basically, supervised learning is performed
by using prior knowledge in output values, whereas unsupervised learning does not use
predefined labels. hence, the goal of this is to infer the natural structures within the
dataset. Therefore, the selection of a machine learning algorithm needs to be carefully
evaluated.

Fig. 1 Model Flow Chart

Dataset
The dataset that was used for the logistic regression analysis is available on the Kaggle
website (https://www.kaggle.com). The classification goal of this study is to predict
whether the patient has a risk of future heart disease. The dataset consists of 300 records of
patient’s data and 14 attributes. The data analysis is carried out in Python programming by
using Jupyter Notebook, which is a more flexible and powerful data science application
software.

Logistic Regression Model

Logistic regression is a one of the machine learning classification algorithms for analysing
a dataset in which there are one or more independent variables that determine an outcome
and categorical dependent variable (DV). Linear regression uses output in continuous
numeric whereas logistic regression transforms its output using the logistic sigmoid
function to return a probability value which can then be mapped to two or more discrete
classes. The logistics regression forms three types as below.
a) Binary logistics regression (two possible outcomes in a DV).
b) Multinomial logistics regression (three or more categories in DV without ordering).
c) Ordinal logistics regression (three or more categories in DV with ordering).
Furthermore, logistic regression model uses more complex cost function instead of linear
function. Logistic regression limits the cost function between 0 and 1.

In the formula, σ () = output between 0 and 1 (probability estimate), z = input to the

function and e = base of natural log.
 Figure 2: Logistic Regression

According to the given data set, 1 indicates the high risk of future heart disease and 0
indicates non or no heart risks. The independent variables n in the logistic model as x1, x2,
x3……., xn

Log ( 1−P
P
)=β + β x + β x + β x … … … .+ β x
0 1 1 2 2 3 3 n n

Logistic regression achieves this by taking the log odds of the event ln(P/1−P), where P is
the probability of event which is risk of heart disease. Therefore, P always lies between 0
and 1.

1.1: Objectives
This report aims to lead the innovation in developing an advanced machine learning model
specifically designed for predicting heart disease, with a primary emphasis on early
detection and thorough risk assessment. The key objectives include utilizing a diverse
range of advanced algorithms. Through the implementation of sophisticated feature
selection techniques.
The designed model is envisioned to excel in early detection and risk stratification,
leveraging a diverse array of patient data. By meticulously analysing various datasets, the
goal is to categorise individuals based on their susceptibility to cardiovascular
complications. A pivotal aspect of this endeavour is optimising the model for clinical
applicability, ensuring seamless integration into existing healthcare workflows, and
thereby providing healthcare professionals with actionable insights for informed decision-
making.
To make sure the model is trustworthy and follows the rules, the report focuses on doing
thorough checks. This includes testing it a lot using different datasets and making sure it
meets ethical and privacy standards.
The aim of this report is to help everyone in the scientific community better understand
how to predict heart disease. By working together and making progress in this area, this
helps the health of people who are at risk of heart disease.

1.2: Methodology
The primary goal of developing this approach was to forecast the likelihood of developing
heart disease. To train our system, we used a variety of feature selection strategies,
including backward elimination and logistic regression, as a machine learning approach
The Kaggle dataset, which has 1026 observations, forecasts the likelihood that a patient
has heart disease or not. Here, we used SK Learn software to predict heart disease using
the patient data provided. With the collected data, pre-processing and loading were carried
out. The preparation procedure comprises deleting the main error and any superfluous data
from the database. This technique is also used to find missing data in a database. Then,
utilizing feature selection, the data pertinent to the prognosis of cardiac illnesses is
extracted.

Data Acquisition and Preprocessing:

Choosing a reliable dataset like Cleveland Heart Disease dataset from Kaggle.com. This
dataset contains various patient attributes and a binary label indicating the presence or
absence of heart disease. Handle missing values through imputation techniques, address
outliers, and encode categorical variables. Analysing the distribution of features,
identifying correlations, and visualizing relationships between features and the target
variable. This helps in feature selection and understanding data patterns.

Figure 3: Dataset distribution

Feature Engineering and Selection:

Feature selection Uses techniques like correlation analysis, chi-square test, or feature
importance methods to select relevant features that contribute significantly to predicting
heart disease. This reduces model complexity and improves interpretability. Feature
transformation applies necessary transformations like scaling numerical features or
creating interaction terms between features to capture complex relationships.

Figure 4: Exploratory data analysis (EDA)

Model Training and Evaluation:

Dividing the pre-processed data into training and testing sets (e.g., 70-30 split). The
training set is used to build the model, and the testing set is used for unbiased evaluation.
Training a logistic regression model on the training set, optimizing its parameters to
minimize the loss function (e.g., binary cross-entropy). Regularization techniques like L1
or L2 can be used to prevent overfitting. Evaluating the model's performance on the testing
set using metrics like accuracy, precision, recall, F1-score. Analysing the confusion matrix
to understand the model's strengths and weaknesses in classifying different types of cases.
Possible Outcomes
The confusion matrix and the extracted values represent the outcomes of a binary classification
model. Let's interpret these outcomes:
Assuming the confusion matrix looks like this:
[[True Negatives False Positives] [False Negatives True Positives]]

Figure 5: Confusion matrix

True Negatives (tn): The number of instances that were correctly predicted as negative (no heart
disease). These are cases where the model correctly predicted that individuals do not have heart
disease, and they indeed do not have it.

False Positives (fp): The number of instances that were incorrectly predicted as positive
(predicted as having heart disease, but they do not). These are cases where the model incorrectly
predicted that individuals have heart disease, but they do not actually have it.

False Negatives (fn): The number of instances that were incorrectly predicted as negative
(predicted as not having heart disease, but they do). These are cases where the model incorrectly
predicted that individuals do not have heart disease, but they do have it.

True Positives (tp): The number of instances that were correctly predicted as positive (having
heart disease). These are cases where the model correctly predicted that individuals have heart
disease, and they indeed have it.