Prediction of Autism Spectrum Disorder using AI

and Machine Learning

2024 18th International Conference on Ubiquitous Information Management and Communication (IMCOM) | 979-8-3503-3101-1/24/$31.00 ©2024 IEEE | DOI: 10.1109/IMCOM60618.2024.10418312

Eiad Yafi Megat Farez Zuhairi

Ritu Chauhan Khushi Mehta Malaysian Institute of Information
Artificial Intelligence and IoT Amity Institute of Biotechnology Faculty of Engineering &
Information Technology Technology
lab, Center for Computational Amity University University Kuala Lumpur
Biology and Bioinformatics UP, India University of Technology Sydney
Sydney, Australia Kuala Lumpur, Malaysia
Amity University [email protected] [email protected]
UP, India [email protected]
[email protected]

Abstract— Knowing the likelihood for a generational shift in the predict issues before they occur, and handle a variety of basic
scientific exploration of digital data which has extensively outgrown challenges. Hence, the current benefits of AI have eventually
society with tons of data. Moreover, the scientific community has attracted several application domains to recognize diverse AI based
evolved and outgrown themselves with Artificial Intelligence based algorithms to discover knowledge from large scale databases.[3]
technology to compute power and detect hidden features and In addition, AI based techniques have the capabilities to handle the
information from outgrown digital databases. These advanced complexity of data without human interaction, where the focus is
systems have the capability to evaluate enormous amounts of data to retrieve patterns and information. [1][4]. The broad area of
generated with unprecedent speed, which can detect valuable insights
research where it embarks to other technology such as machine
from the data. In addition, the current study of approach focuses on
learning, deep learning, human cognitive and many others.
the application of AI with healthcare databases to detect the hidden
information from autism spectrum disorder (ASD). In this research
different branches of AI are functional to other application domains
we have various algorithmic predictions for Autism Spectrum such as business, education, banking retail, and healthcare services
Disorder (ASD) by combining Machine Learning with the dataset to detect hidden information and patterns for discovering
provided. The proposed approach is based on the predictive method knowledge among the databases [5]. Investigating various uses of
where the data preprocessing and data evaluation models are used. AI, our analysis found that modern technology may assist one
The study of approach focuses on ASD with several machine advance their career in the healthcare field. The evolution of AI has
Learning algorithms such as, SVM, KNN, Random Forest, Decision made its way in the Healthcare which have shown various
Tree to measure the accuracy and precision among the data and advantageous and righteous tasks such as: Amazing and precise
determine the hidden information and patterns. The current research diagnosis, successful surgeries, medical care for uncommon
scenario investigated concurrent parameters which are dependable disorders, Chatbots, and health care record monitoring[3][6][24-
features such as Class\ASD Traits, Sex, Jaundice scores and Q chat- 25]. Moreover, the AI has paved it way into the diagnosis of
10 questions by which we have come to a conclusion which treating many neurological disorders which includes OCD, ADHD,
corroborated the current analysis with an accuracy Score of 74% for and moreover ASD. Autism Spectrum Disorder (ASD), which is a
Random Forest, KNN: 71%, SVM: 70.6% and Decision Tree: 69.1% neurodevelopmental disorder resulting in limiting interactions and
stating whether the individual has the Autism characteristics visible deficit in social communications. Childrens may transmit emotions
or not using various parametric analysis through data visualization by the colour of their eyes, the placement of their lips and pupils,
and analyzing results by this approach.
and the pitch of their speech [7]. Childrens with this sort of issue
Keywords— Artificial intelligence (AI), machine learning (ML),
are more willing to listen to treatment than an actual human
healthcare, Autism Spectrum Disorder
psychologist. Toddlers and teenagers with this disorder should be
I. INTRODUCTION
diagnosed as early as possible, and with the use of correct
The use of Artificial Intelligence (AI) has paved its way in the
technology, we can cure and functionalise the symptoms
modern era to remove the ineffectual and tedious tasks by applying
beforehand. For instance, a specific sort of robot has been
automated tool and technology. Hence, existing usage of AI has
developed to address this issue which are termed “Hard Bank
considerably applied to transform the data into a useful or a quick
Robots”, approximately two feet tall where Adolescents can utilise
knowledge system to detect information from large scale
new abilities in a way which is more comparable to working with a
databases.[1] Hereafter, knowledge accessed while the usage of AI
person, but that is less sophisticated and dynamic, with an
can be utilized for real time monitoring decision making. So, AI
automated device.[8]. ASD prediction relies heavily on ML
can also be assessed as Intelligent automation system which can
techniques and algorithms which calculates and evaluate various
automate tasks, improve prediction, and add a variety of convenient
parameters to conclude the obvious.[9]. With the goal to assess the
solutions to common problems for effective and efficient patterns
precision and dependability of ASD forecasting models,
detection from structured and unstructured databases.[2] We can
performance assessment measures, double validation techniques,
say that AI can support with intricate problem-solving activities,
and model selection procedures are discussed. [10]. It has

979-8-3503-3101-1/24/$31.00 ©2024 IEEE

Authorized licensed use limited to: Zhejiang University. Downloaded on November 05,2024 at 17:11:58 UTC from IEEE Xplore. Restrictions apply.
challenging to deploy AI technologies in the healthcare industry A study found in 1967, focused on the ASD prediction from eye-
that have an influence on clinical treatment for a few domain- tracking technology data. This research project focuses on eye-
specific reasons. Construction timelines have been lengthier than in tracking data analysis to identify ASD by analysing patterns of
other disciplines because of issues with algorithm assessment, attention span. Recurrent neural networks, or RNNs for short,
model maintenance, integration with EHR or hospitals data centres, were implemented by scientists to develop a deep learning
as well as confidentiality and fairness problems. [11]. The current method that accurately predicted ASD by illuminating temporal
research investigates methods for determining feature importance correlations in visual tracking loops. [16]. The current
and deriving significant knowledge from predictive machine investigation aims, Longitudinal Prediction of Autism Spectrum
learning (ML) models. Specifically, we have employed Support Disorder via Early Behavioural Features. The scientists captured
Vector Machine (SVM), K-Nearest Neighbors (KNN), Random neural circuits and accurately predicted the consequences of ASD
Forest and Decision Tree for machine learning methods to evaluate using longitudinal ML techniques, such as mixed-effects
and predict the disorder. The suggested strategy is based on a algorithms and secret Markov models , the study was done to
predictive method that employs models for data preparation and forecast ASD using early developmental markers acquired
assessment. The study's methodology centres on ASD and employs throughout time[17].
a number of machine learning algorithms, including SVM, KNN,
Random Forest, and Decision Tree, to assess the data's accuracy
and precision, find patterns in the data, and detect any hidden
information. We came to a conclusion that supported the current
analysis with an accuracy score of 74% for Random Forest, KNN: III. RESEARCH METHODOLOGY
71%, SVM: 70.6%, and Decision Tree: 69.1% by investigating A. DATASET
concurrent parameters that are reliable features such as ClassASD
Traits, Sex, Jaundice scores, and Q chat–10 questions assessing
whether an individual has Autism symptoms by doing various
parametric analyses using data visualisation and then evaluating the
outcomes.

II. LITERATURE REVIEW

Researchers are using an incorporating MR imaging data, to offer

a prompt diagnosis of autism spectrum disease. A deep learning-
powered approach for Autism identification employing structural
MRI data was presented in this study. The researchers achieved
exceptional precision in early diagnosis by using an architecture
based on convolutional neural networks (CNN) with an accuracy
score of 71.62% [12]. Research done in this field of study over
a decade ago, focused on using Data Mining and Natural
Language Processing to Predict Autism Spectrum Disorder. In
this research, the use of NLP (natural language processing) Fig 1. Illustrates Q-Chat 10 questions for Autism.
techniques for ASD detection was examined. The scientists
developed an NLP-based framework for assessing language Fig1 illustrates the Q-Chat 10 questions, where we can see ten
patterns in healthcare documentation and evaluations, enabling different simple questions are being assessed and considered on
them to precisely predict ASD in early adolescents.[10],[13] An which we can determine the symptoms of the Autism.
ensemble-based learning method is employed to identify autism
spectrum diseases using EEG data. Researchers proposed an
ensemble learning framework that used numerous ML The data entails cleaning the data, removing any missing values
techniques, such as decision trees and support vector machines, or throughout the purposes of this research, ten personality
showing an accuracy score of 78% and 68.23% , in an attempt characteristics (Q-Chat-10) have been recorded together with
to increase the accuracy of the diagnosis of ASD based on the other individual traits that have been shown to be valuable and
information from the EEG. The use of electroencephalogram helpful in separating those with autism spectrum disorders from
(EEG) signals for ASD prediction were the focus of this work. norms in neurological and cognitive sciences. Autism spectrum
[14].Making use of intraoperative nuclear medicine and machine disorder (ASD), a cognitive and psychiatric illness linked to high
learning , the researchers predicted the severity of an autism healthcare expenses, can be considerably reduced with early
spectrum disorder. This study aimed at anticipating the intensity identification.[18][11]. Because of the dramatic increase in ASD
of ASD using fMRI (functional magnetic resonance imaging) cases throughout the world, data sets relevant to behavioural
data. The study found patterns in brain connections associated to aspects are necessary. As a result, a quick and simple ASD test
the severity of ASD using machine learning techniques, such as will soon be available to aid medical practitioners and inform
logistic models and decision-making algorithms, and provided patients whether they should seek a formal clinical diagnosis
strong predictive ability. [15]. Because of the dramatic increase in ASD cases throughout the
world, data sets relevant to behavioural aspects are
necessary.[19] . A1-A10: Q-Chat 10 items that feature question

Authorized licensed use limited to: Zhejiang University. Downloaded on November 05,2024 at 17:11:58 UTC from IEEE Xplore. Restrictions apply.
 and probable answers. The dataset converts the value of the that are used in the computational arrangement shown as a form of
"Always, Usually, Occasionally, Rarely, and Never" entries to equation in Eq.2.
"1" or "0". If the response was Sometimes / Rarely / Never, the
question (A1-A9) is given "1" for questions 1 through 9 in Q- dist(x,z)=(∑r=1d|xr−zr|p)1/p (2)
chat-10. Question 10 (A10), on the other hand, earns a "1"
whether the answer is Always, Usually, or Occasionally. Add the • Random Forest
points for each of the 10 questions if the user received more than Each tree in a random forest relies upon the values of a random
three. If the child obtains a score more than 3 (Q-chat-10- score), vector that was sampled independently and with the same
he or she has a possible ASD characteristic; otherwise, no ASD distribution for all the trees in the forest. As the variety of trees in
symptoms are seen. The given parameters are: A1, A2,
a forest increases, the generalization error integrates as towards a
A3,A4,A5,A6,A7,A8,A9,A10, Age_mons, Qchat-10-
limit. The strength of each individual tree and the correlation
score,Sex,Jaundice,Family_mem_ASD[17][20]
between them determine the degree of generalization error of a
B. Pre-Processing Data forest of branch analysers.
The machine learning model performed was impacted by missing • Decision Tree
values, duplicate entries, and outliers. As a result, before training
the model, the dataset was cleaned. The values that are missing Decision Tree is a Guided learning strategy that may be used to
were replaced with the proper value or eliminated entirely. solve regression as well as classification issues, however, several
Duplicate entries got removed to ensure that each data point is machine learning algorithms for decision trees operate by
unique. Outliers were removed as well to prevent them from repeatedly dividing one of the sections into two. Only the leaf
skewing the results.The dataset was separated into sets for training nodes can be divided; each portion may be divided a maximum of
and testing in order to prepare it for automated learning. This split once. Every portion may be split numerous times in switching
is critical for appropriately assessing the model's efficacy. In order decision-making trees in general[23].
to accomplish this, we used the scikit-learn library's
train_test_split method [21]. The characteristics variables, written IV. RESULT
as x, encompass the first 15 categories of the dataset, whereas the By studying the dataset, we can conclude the relations in between
target variable, denoted as y, corresponds to the 16th column, the given parameters in this heatmap. Family_mem_with_ASD
which is labeled 'Class/ASD Traits.' The division was set up using and various ASD trait variables (A1, A2, A3, A6, A7, A8, A9, and
a 75%-25% ratio, with 75% of information designated for training A10) supports a genetic or family effect on the existence of ASD
and 25% for testing. In addition, we established a random_state to features. heatmap is best to visualise the data analysis.
0 to assure the split's reproducibility.
A. A Explorative Data Analysis
C. Algorithms Used By studying the dataset, we determined the relations in between
• Support Vector Machine (SVM) the given parameters in this heatmap. Family_mem_with_ASD
and various ASD trait variables supports a genetic or family effect
SVM is one of the most popular Machine Learning and Data
on the existence of ASD features. This conclusion supports prior
Mining algorithm which is used for classification problems
studies demonstrating the genetic basis of ASD. Understanding
chooses the extreme points/vectors that help in creating the
the relationship between family history and ASD features can
hyperplane. These severe circumstances are known as support
help with genetic investigations and give insights into inheritance
vectors, and the technique is known as the Support Vector
patterns and potential risk factors for ASD. Overall, when it
Machine. If the total quantity of features entered is two, the hyper-
comes to finding the positive and negative correlation between
plane is simply a line. When the total quantity of input
the distinct parameters, the heatmap is best to visualise the data
characteristics reach-es three, the hyperplane transforms into a
analysis.
two-dimensional plane shown in Eq.1. It gets tough to picture
when a list of characteristics reaches three.[22]
w·X+b=0 (1)
Positive numbers indicate one class, while negative values suggest
the other, and the sign of "w X + b" defines the class.
• K-Nearest Neighbour’s (KNN)
The k-nearest neighbour’s algorithm, known as KNN is a
supervised learning classifier, uses proximity to make
classifications or predictions about the grouping of an individual
data point. While it can be used for either regression or
classification problems, it is typically used as a classification
algorithm That operates on the idea that the data that are closest to
an evident point are the most " analogous " analyses in a collection Fig 2. Illustrates the Heatmap Analysis of the given parameters which can
of data. By selecting K, we can classify undesirable points help us detect the disorder
according to what is known about the closest continuing points,
and the user can also select the total number of close-by appraisals

Authorized licensed use limited to: Zhejiang University. Downloaded on November 05,2024 at 17:11:58 UTC from IEEE Xplore. Restrictions apply.
Fig. 2 illustrates the correlations between the attributes provided central trend and dispersion. This information is helpful for
to the dataset. The correlation between different parameters comparing the score between the two-gender provided and for
including A1, A2, A3, A4, A5, A6, A7, A8, A9 ,A10, spotting probable differences or patterns within these groups.
Age_mons,Qchat-10-score,Sex,Jaundice,Family_mem_ASD. The plot is displayed using the plt.show() function for your
examination and evaluation. Understanding the dataset's
In this lieu of ASD prediction and screening, researchers and evolving patterns as well as distribution is greatly aided by this
medical practitioners have better target therapies and care for graphical depiction. By displaying the frequency of various age
those who suffer from the ASD by identifying paticular values, we encounter trends, outliers, or frequent age groupings
behaviors and traits associated with ASD with this correlation. in the dataset.
The correlations may vary among datasets or
B Result and Evaluation
groups.ThePpositiverelrelationshipbetbetweenFamily_mem_wit
h_ASD and various ASD trait variables (A1, A2, A3, A6, A7, To examine and forecast the outcomes of the values obtained
A8, A9, and A10) supports a genetic effect the probability from the experiments, we have employed a variety of machine
supporting ASD traits. learning algorithms measures, including KNN, Random Forest,
Decision Tree, and SVM (Support Vector Machine) for assessing
autism spectrum disorder (ASD) and determining outcome
predictions for both the qualitative and quantitative viewpoints.
Class zero had a score of 0.58, implying that 58 percent of
instances that had been designated as class zero really
represented actual negatives. 91% of occurrences classified as
class 1 were true positives, according to the degree of certainty
for class 1, which is 0.91. The recall score represents the ML
system ability to recognize all positive cases. 80% of true
negative were properly detected because Class 0 has a recall of
0.80Recall forclClassok1isisal0.78, meaning roughly 8% of real
Fig 3. Illustrates a Count plot analysis to evaluate the age based on the ASD positives were correctly identified.
prediction.
Table1. Consists of accuracy scores of different algorithms

Figure 3 shows a bar plot that acts as a frequency meter to show Model Name Accuracy Precision Recall
how the data in the 'Age_mons' column are distributed. By
displaying different age values and their accompanying SVM 0.7 0.7 0.7
frequencies, this visualization efficiently communicates the age KNN 0.71 0.71 0.71
distribution within a particular age group. The x-axis has been Random Forest 0.74 0.74 0.74
imaginatively labelled "Age_mons”(depicting individual’s age)
and the y-axis has been labelled "count" to help in the Decision Tree 0.69 0.69 0.69
comprehension of this visualization. Here the age is shown on
the x axis(positive), and y axis is being assigned to the frequency
Table 1 entails various algorithmic accuracy scores, precision
which the ranges are assigned to.
values and recall values for KNN, Random Forest, Decision Tree
and SVM. The algorithms have showed the accuracy scores of
74% for Random Forest, 70.6% for SVM, 69.1% for Decision
Tree and, 71% for KNN, following with a great precision score
as well for the dataset evaluation. Class 0 has an F1-score of 0.67,
whereas Class 1 has an F1-score of 0.84. Class 0 has 74
supporters, whereas we can see Class 1 has 190 supporters. The
macro-average and weighted-average scores provide such
assessment of the machine learning (ML) techniques accuracy
over subclasses. The weighted-average F1-score is 0.79, whereas
it is visible that the macro-average is 0.76. The confusion matrix
throws more light on the workings of the ML algorithms: [59 15]
[42 148]]. Depending on the confusion matrix, 59 of the 74 cases
in class 0 were accurately expected.
Fig 4. Illustrates a violin plot to visualize the Qchat-10-Score distribution by
Sex
Table2. Displaying Precision and Recall matrices scores for different algorithm
In figure4., the y-axis displays the Qchat-10-Score, while the x- used
axis depicts gender ('Sex'). The headline "Qchat-10-Score Precision Recall
Variance by Class," which specifies "Class" while the x-axis ML Models Average Average Average Average
denotes Sex in the violin plot. The violin plot is used to illustrate (Macro) (Weighted) (Macro) (Weighted)
the 'Qchat-10-Score' values so that you can see the way they are SVM 0.7 0.71 0.7 0.7
divided across the various genders ('Sex'). Insights into the KNN 0.71 0.71 0.72 0.69
distribution's shape are provided by violin plots, which show a

Authorized licensed use limited to: Zhejiang University. Downloaded on November 05,2024 at 17:11:58 UTC from IEEE Xplore. Restrictions apply.
 Random m
0.74 0.74 0.74 0.73 to their distribution. With an AUC score of 0.61, Random Forest
Forest is the most powerful classifier among the examined models. With
Decision Tree 0.69 0.69 0.69 0.69 AUC values of 0.56, KNN and Decision Tree perform
sufficiently. In comparison, the SVM's AUC of 0.5 denotes
Table 2 displays both precision and recall scores for four machine classification accuracy equal to chance. These ROC analysis
learning models, with the Random Forest model showing the best results offer insightful guidance for model selection and
and most even result across both macro and weighted averages. highlight the possibility for further model optimization and
Both macro and weighted accuracy and recall for the Support tailoring to improve classification quality.
Vector Machine (SVM) are routinely around 0.70, demonstrating
V. FUTURE WORK AND DISCUSSION
equal effectiveness across several classes. Particularly in macro
recall, where it reaches 0.72, the K-Nearest Neighbor’s (KNN) The evaluation and therapy of Autism Spectrum Disorder (ASD)
model has a little greater accuracy and recall. The Random Forest holds tremendous potential for future study in the fields of
algorithm shines out with good recall and accuracy scores of 0.74 artificial intelligence and machine learning. First, further
across both macro and weighted averages, emphasizing its research into sophisticated deep learning models, such as
outstanding categorization skills. On the contrary, the Decision convolutional neural networks (CNNs) and recurrent neural
Tree algorithm keeps its constant but slightly lower accuracy and networks (RNNs), might be done to improve accuracy. These
recall scores at 0.69. models are likely to increase the accuracy of ASD diagnosis
since they are recognized for their prowess in processing
Fig 5. Illustrates the ROC curve examining different algorithms.
complicated data, such as medical pictures and time series data.
Another crucial route is the integration of multimodal data. A
more thorough knowledge of ASD may be attained by combining
data from many sources, including medical pictures, EEG
signals, and behavioral features. In order to provide more precise
and nuanced assessments, further studies must emphasize the
integration of these many data sets. Real-time monitoring and
intervention systems also require consideration. The creation of
AI-based solutions for continuing care and individualized care
plans for people with ASD is essential. As healthcare AI
advances raise concerns about data security and patient
confidentiality, ethical and privacy problems must continue to be
at the forefront of future effort. It is essential to have AI systems
that are trustworthy, transparent, and follow moral rules. The
Fig.5 illustrates the ROC Curve analysis graph, in which we can reliability and effectiveness of AI-based diagnostic tools for
see various algorithms having dissimilar ROC values. The ASD must be confirmed by clinical validation and integration
algorithms scores are pointed on the True Positive Rate and False studies, which call for partnerships with healthcare organizations
Positive Rate. and extensive clinical trials. The focus should also move to early
intervention and individualized treatment planning, which will
Using these values, we have plotted AU-ROC curve for the make use of AI to adapt treatment strategies to specific
algorithms we used to predict the results: SVM, KNN, Random demographics of patients.
Forest, Decision Tree. Each algorithm is trained and tested.
Additionally, the legend provides the AUC values for each The deployment of AI-based diagnostic tools in healthcare
algorithm, which allows an instantaneous comparison of their settings should be streamlined and regulatory issues should be
performance. Greater differentiation is shown by higher AUC addressed in future research, guaranteeing fulfillment of
values, which suggests that the algorithm is better able to regulatory requirements while benefiting patients and medical
distinguish between those with ASD and those without it. Using professionals.
the provided code, a Receiver Operator Characteristic (ROC)
curve is produced, a useful visual aid for evaluating the
effectiveness of predictive binary models. The balance within the CONCLUSION
true positive rate (TPR) and the false positive rate (FPR) is shown Advances in science-driven artificial intelligence (AI) and
in this scenario using the ROC curve for four distinct machine machine learning have implications that go beyond processing
learning models: K-Nearest Neighbor’s (KNN), Decision Tree, power in terms of the possibility of a generational change in
Random Forest, and Support Vector Machine (SVM). With AUC scientific research at DOE Laboratories. It needs a wide
scores of 0.56, KNN and Decision Tree represented by the green variety of computing infrastructures, such as real-time
and blue dotted lines, respectively perform somewhat better than tracking, collaborative tools, and statistical analysis. These
random guessing. These models' ROC curves show a harmony state-of-the-art technologies have the power to interpret vast
between accurately comprehending positive instances and amounts of medical data, provide meaningful analysis, and
reducing false positives. Exhibiting a mean AUC of 0.61, raise the ability to diagnose and treatment the precision of
Random Forest stands out with a stronger capacity to health care providers. Integrating AI and ML into healthcare
discriminate between positive and negative cases (shown by the organizations does, however, come with certain difficulties,
purple dashed line). TPR and FPR are better traded off according including data security, regulatory compliance, interpreting

Authorized licensed use limited to: Zhejiang University. Downloaded on November 05,2024 at 17:11:58 UTC from IEEE Xplore. Restrictions apply.
 the results of sophisticated algorithms, and ethical issues. It quantitative checklist in 1000 cases and 3000 controls", J. Amer. Acad. Child
needs a wide variety of computing infrastructures, such as Adolescent Psychiatry, vol. 51, no. 2, pp. 202-212, 2012.
real-time tracking, collaborative tools, and statistical analysis. [7] J. Peral, D. Gil, S. Rotbei, S. Amador, M. Guerrero, and H. Moradi, “A
machine learning and integration based architecture for cognitive disorder
The ASD, with other Neurological and cognitive disorders can detection used for early autism screening,” Electronics, vol. 9, no. 3, 2020.
be treated with a thorough mechanism including AI and ML [8] Md Delowar Hossain, Muhammad Ashad Kabir, Adnan Anwar and Md
for the beneficial use curing hundreds and thousands of Zahidul Islam, "Detecting Autism Spectrum Disorder using Machine
patients with medical imaging and research analysis. This Learning-An Experimental Analysis on Toddler", Child and Adult Adults,
study explores the use of machine learning algorithms in 2020
artificial intelligence (AI) applications for the prediction of [9] N. Gabbay-Dizdar, M. Ilan, G. Meiri, M. Faroy, A. Michaelovski, H. Flusser,
Autism Spectrum Disorder (ASD). In order to effectively treat et al., "Early diagnosis of autism in the community is associated with marked
improvement in social symptoms within 1-2 years" in Autism, SAGE
ASD, the study emphasizes the value of early detection and Publications Ltd, vol. 26, no. 6, pp. 1353-1363, Aug. 2022
treatment, while also stressing the potential of AI and machine
[10] M. Mehmood, M. Rizwan, M. Gregus ml and S. Abbas, "Machine Learning
learning to assist in this procedure. Assisted Cervical Cancer Detection", Frontiers in Public Health, vol. 9, 2021.
Exploratory data analysis revealed relationships between [11] Chanel, G.; Pichon, S.; Conty, L.; Berthoz, S.; Chevallier, C. and Grèzes, J.
(2016). “Classification of autistic individu-als and controls using cross-task
family history and specific ASD features, supporting the characteri-zation of fMRI activity”. NeuroImage: Clin-ical
genetic basis of ASD and supporting earlier findings. Dataset
[12] Abdullah, A. A.; Rijal, S., and Dash, S. R. (2019). “Evaluation on Machine
trends were highlighted using the heatmap , age distribution Learning Algorithms for Classification of Autism Spectrum Disorder
and Violin plot visualization. The Random Forest model (ASD)”. In Journal of Physics: Conference Series, 1372(1), p. 012052. IOP
continuously performed better in the evaluation of precision Publishing
and recall measures, demonstrating the accuracy of its ASD [13] Heinsfeld, A. S.; Franco, A. R.; Craddock, R. C.; Buchweitz, A. and
categorization. These findings were confirmed by ROC curve Meneguzzi, F. (2018). “Identification of autism spectrum disorder using deep
analysis, with the Random Forest model attaining the greatest learning and the ABIDE dataset”. NeuroImage: Clinical
AUC, demonstrating its superior capacity to differentiate [14] H. S. Nogay and H. Adeli, "Machine learning (ML) for the diagnosis of
autism spectrum disorder (ASD) using brain imaging", Reviews in the
between positive and negative ASD patients. A number of Neurosciences, vol. 31, no. 8, pp. 825-841, 2020.
machine learning models, including Support Vector Machine [15] S. Batsakis, M. Adamou, I. Tachmazidis, S. Jones, S. Titarenko, G. Antoniou,
(SVM), K-Nearest Neighbor’s (KNN), Random Forest, and et al., "Data-driven decision support for adult autism diagnosis using machine
Decision Tree, were employed for forecasting ASD through learning", Digital, vol. 2, no. 2, pp. 224-243, 2022.
the study of a dataset containing different metrics and ASD [16] J Samuel, A. (1967). Some studies in machine learning using the game of
symptoms. In this Research, the machine learning algorithms checkers II: Recent progress. IBM J. Research and Development .
have shown the classification findings and an accuracy score [17] J. Devlin, M.-W. Chang, K. Lee and K. Toutanova, "BERT: Pre-training of
of 74%, 70.1%, 69.1%, and 71%, confirming that the child deep bidirectional transformers for language understanding" in Proceedings
have Autism or not since the traits measured and evaluated, of the 2019 Conference of the North American Chapter of the Association for
Computational Linguistics: Human Language Technologies Volume 1 (Long
where the Random Forest having the best accuracy score from and Short Papers), Minneapolis, Minnesota:Association for Computational
all other. This study highlights how AI and machine learning Linguistics, pp. 4171-4186, Jun. 2019.
might help with ASD early detection and therapy. Further [18] Benjamin Wingfield, Shane Miller, Pratheepan Yogarajah, Dermot Kerr,
study and model improvement have the potential to improve Bryan Gardiner, Sudarshi Seneviratne, et al., "A predictive model for
ASD prediction, eventually helping those who are impacted paediatric autism screening", Health informatics journal, vol. 26, no. 4, pp.
by this illness. 2538-2553, 2020
[19] X. Guo et al., "Diagnosing Autism Spectrum Disorder from Brain Resting-
State Functional Connectivity Patterns Using a Deep Neural Network with a
Novel Feature Selection Method", Front Neuro sci, vol. 11, pp. 460, 2017
REFERENCES [20] A. Liaw, M. Wiener et al., "Classification and regression by randomforest",
R news, vol. 2, no. 3, pp. 18-22, 2002.
[1] Taban Eslami, Fahad Almuqhim, Joseph s. Raiker and Fahad Saeed, Machine [21] F. N. Buyukoflaz and A. Ozturk, "Early autism diagnosis of children with
Learning Methods for Diagnosing Autism Spectrum Disorder and Attention- machine learning algorithms", 2018 26th Signal Processing and
Deficit/Hyperactivity Disorder Using Functional and Structural MRI: A Communications Applications Conference (SIU), pp. 1-4, 2018.
Survey, 2021 [22] Tabtah, F. (2017). Autism Spectrum Disorder Screening: Machine Learning
[2] R. Sreedasyam, A. Rao, N. Sachidanandan, N. Sampath and S. K. Vasudevan, Adaptation and DSM-5 Fulfillment. Proceedings of the 1st International
"Aarya—A kinesthetic companion for children with autism spectrum Conference on Medical and Health Informatics 2017, pp.1-6. Taichung City,
disorder", J. Intell. Fuzzy Syst., vol. 32, no. 4, pp. 2971-2976, Mar. 2017. Taiwan, ACM.
[3] K.-F. Kollias, C. K. Syriopoulou-Delli, P. Sarigiannidis and G. F. Fragulis, [23] A. S. Albahri, R. A. Hamid, O. S. Albahri, and A. A. Zaidan, “Detection-
"The contribution of machine learning and eye-tracking technology in autism based prioritisation: framework of multi-laboratory characteristics for
spectrum disorder research: A systematic review", Electronics, vol. 10, no. asymptomatic COVID-19 carriers based on integrated entropy–TOPSIS
23, pp. 2982, Nov. 2021. methods,” Artificial Intelligence in Medicine, vol. 111, article 101983, 2021.
[4] M. D. Hossain, M. A. Kabir, A. Anwar and M. Z. Islam, "Detecting autism [24] R Chauhan,., E Yafi,. Applicability of classifier to discovery knowledge for
spectrum disorder using machine learning techniques", Health Inf. Sci. Syst., future prediction modelling. J Ambient Intell Human Comput 14, 10347–
vol. 9, no. 1, pp. 1-13, Dec. 2021. 10362 (2023). https://doi.org/10.1007/s12652-022-03694-3
[5] S. P. Abirami, G. Kousalya and R. Karthick, "Identification and exploration [25] R. Chauhan, N. Gola and E. Yafi, "An Analytical Approach to Predict the
of facial expression in children with ASD in a contact less environment", J. Cardio Vascular Disorder," 2023 17th International Conference on
Intell. Fuzzy Syst., vol. 36, no. 3, pp. 2033-2042, Mar. 2019. Ubiquitous Information Management and Communication (IMCOM), Seoul,
Korea, Republic of, 2023, pp. 1-8, doi:
[6] C. Allison, B. Auyeung and S. Baron-Cohen, "Toward brief ‘red flags’ for 10.1109/IMCOM56909.2023.10035581.
autism screening: The short autism spectrum quotient and the short

Authorized licensed use limited to: Zhejiang University. Downloaded on November 05,2024 at 17:11:58 UTC from IEEE Xplore. Restrictions apply.
 Meneguzzi, F. (2018). “Identification of autism spectrum disorder using deep
learning and the ABIDE dataset”. NeuroImage: Clinical
[14] H. S. Nogay and H. Adeli, "Machine learning (ML) for the diagnosis of
autism spectrum disorder (ASD) using brain imaging", Reviews in the
Neurosciences, vol. 31, no. 8, pp. 825-841, 2020.
[15] S. Batsakis, M. Adamou, I. Tachmazidis, S. Jones, S. Titarenko, G. Antoniou,
et al., "Data-driven decision support for adult autism diagnosis using machine
learning", Digital, vol. 2, no. 2, pp. 224-243, 2022.
[16] J Samuel, A. (1967). Some studies in machine learning using the game of
checkers II: Recent progress. IBM J. Research and Development .
[17] J. Devlin, M.-W. Chang, K. Lee and K. Toutanova, "BERT: Pre-training of
deep bidirectional transformers for language understanding" in Proceedings
of the 2019 Conference of the North American Chapter of the Association for
Computational Linguistics: Human Language Technologies Volume 1 (Long
and Short Papers), Minneapolis, Minnesota:Association for Computational
Linguistics, pp. 4171-4186, Jun. 2019.
[18] Benjamin Wingfield, Shane Miller, Pratheepan Yogarajah, Dermot Kerr,
Bryan Gardiner, Sudarshi Seneviratne, et al., "A predictive model for
paediatric autism screening", Health informatics journal, vol. 26, no. 4, pp.
2538-2553, 2020
[19] X. Guo et al., "Diagnosing Autism Spectrum Disorder from Brain Resting-
State Functional Connectivity Patterns Using a Deep Neural Network with a
Novel Feature Selection Method", Front Neuro sci, vol. 11, pp. 460, 2017
[20] A. Liaw, M. Wiener et al., "Classification and regression by randomforest",
R news, vol. 2, no. 3, pp. 18-22, 2002.
[21] F. N. Buyukoflaz and A. Ozturk, "Early autism diagnosis of children with
machine learning algorithms", 2018 26th Signal Processing and
Communications Applications Conference (SIU), pp. 1-4, 2018.
[22] Tabtah, F. (2017). Autism Spectrum Disorder Screening: Machine Learning
Adaptation and DSM-5 Fulfillment. Proceedings of the 1st International
Conference on Medical and Health Informatics 2017, pp.1-6. Taichung City,
Taiwan, ACM.
[23] A. S. Albahri, R. A. Hamid, O. S. Albahri, and A. A. Zaidan, “Detection-
based prioritisation: framework of multi-laboratory characteristics for
asymptomatic COVID-19 carriers based on integrated entropy–TOPSIS
methods,” Artificial Intelligence in Medicine, vol. 111, article 101983, 2021.
[24] R Chauhan,., E Yafi,. Applicability of classifier to discovery knowledge for
future prediction modelling. J Ambient Intell Human Comput 14, 10347–
10362 (2023). https://doi.org/10.1007/s12652-022-03694-3
[25] R. Chauhan, N. Gola and E. Yafi, "An Analytical Approach to Predict the
Cardio Vascular Disorder," 2023 17th International Conference on
Ubiquitous Information Management and Communication (IMCOM), Seoul,
Korea, Republic of, 2023, pp. 1-8, doi:
10.1109/IMCOM56909.2023.10035581.

Authorized licensed use limited to: Zhejiang University. Downloaded on November 05,2024 at 17:11:58 UTC from IEEE Xplore. Restrictions apply.