Exploring the Potential of Machine Learning in Healthcare Accuracy

Improvement

SINDHU VEERAMANI1,*, S. M. RAMESH2, B. GOMATHY3

1Dept. of Computer Science and Engineering,

Dr. NGP Institute of Technology,

Coimbatore, Tamil Nadu,

INDIA

2Dept. of Electronics and Communication Engineering,

KPR Institute of Engineering and Technology,

Coimbatore, Tamil Nadu,

INDIA

3Dept. of Computer Science and Business Systems,

Dr. NGP Institute of Technology,

Coimbatore, Tamil Nadu,

INDIA

Abstract: - Machine learning techniques have shown great potential in the medical industry, particularly in the

field of neuroimaging and the identification of neurological illnesses such as Autism Spectrum Disorder (ASD).

By utilizing machine learning algorithms, researchers aim to predict the type of disability and analyze the

predicted variations using different types of predictive models. These predictive models can be trained on

neuroimaging data to identify patterns and markers that are indicative of ASD. By analyzing these patterns,

machine learning algorithms can help in accurately predicting the presence and type of ASD in individuals.

This can be immensely valuable in early diagnosis and intervention, leading to better outcomes for individuals

with ASD. Furthermore, the applications of machine learning in the healthcare industry extend beyond just

prediction. Machine learning algorithms can also be used to analyze large amounts of medical data, identify

trends, and assist in decision-making processes. This can help healthcare professionals in providing more

accurate diagnoses, personalized treatment plans, and improved patient care. It is important to note that the

success and accuracy of machine learning models in the healthcare industry depend on various factors,

including the quality and quantity of data available, the choice of algorithms, and the expertise of the

researchers. Ongoing research and advancements in machine learning techniques hold great promise for

improving the accuracy and effectiveness of medical diagnoses and treatments.

Key-Words: - Autism Diagnostic Observation Schedule (ADOS), Autism Spectrum Disorder (ASD), Autism

Diagnostic Interview-Revised (ADI-R), Classification Algorithms, Machine Learning(ML),

Support Vector Machine (SVM), Decision Tree (DT), K Nearest Neighbors (KNN), Naive Bayes

(NB), Logistic Regression (LR), Random Forest Classifier (RFC).

Received: August 28, 2023. Revised: November 22, 2023. Accepted: December 19, 2023. Published: December 31, 2023.

1 Introduction

Autism Spectrum Disorder (a neurological illness) is

an enlightening inability caused by Variances in the

brain. There are specific difficulties with verbal and

gesture communication, social interaction, and

recurring behavior among persons with autism.

Autism spectrum disorder will be acknowledged by

Medical Consultants by analyzing the variation in

people's behavior with the condition based on their

age and level of competence. [1], [2], [ 3], ASD has

a long-term impact on individuals, their families,

and the provision of educational and therapeutic

services. DSM-5 describes human interaction

abilities as (a) deficiencies in teamwork, (b)

irregularities in communication skills, and (c)

problems establishing, sustaining, and

comprehending relationships. In light of problems

with effective interaction and repeating and

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2023.22.42

Sindhu Veeramani, S. M. Ramesh, B. Gomathy

E-ISSN: 2224-2872

374

Volume 22, 2023

restricted behavior, DSM-5 classified ASD into

three severity levels. Level one - Requiring Support,

Lack of social communication causes obvious

impairments if there are no supports in place. Level

two - Requiring Substantial Support, Social

impairments are noticeable even with assistance,

and Significant weaknesses in verbal and nonverbal

social communication abilities level three -

Requiring Very Substantial Support - Ability to

initiate social contacts is severely hampered by

severe inadequacies and there is little reaction to

social advances made by others.

The ADI-R and ADOS are two clinical instruments

and diagnostic approaches that are unfortunately

time-consuming and necessary for the medical

identification of people with autism (ADOS), [4].

Parent interviews are a part of the rigorous,

organized, and structured ADI-R test, which looks

at the child's early life, [5]. ADOS advises the

parent about the child's current social,

communication, and play abilities during a

prearranged play session.

2 Literature Survey

Several kinds of research with the help of machine

learning was done to speed up and enhance the

diagnosis of ASD. Utilizing machine learning,

which had an accuracy of 84%, to categorize the

retinal obsessions of kids with ASD and TD. These

investigations showed that when compared to

established diagnostic measures, machine learning is

more efficient and objective, [6]. The integrating

multidimensional features gives the best degree of

accuracy for depicting the correlation coefficients of

the brain when compared to examining individual

heterogeneous variables, [7]. With an accuracy of

97.6%, The Support Vector Machine technique to

screen for ASD. The limitation of this paper is the

small sample containing 612 autistic cases and 11

non autistic cases, [8]. Sixty-five Social

Impartiality individuals from two thousand nine

hundred and twenty-five individuals with ASD or

ADHD with the help of six machine learning

models. They employed forward feature selection

with minimum sampling, [9]. Five out of the sixty-

five tests, with an efficiency of 96.4%, were

sufficient to differentiate ASD from ADHD. There

was a considerable imbalance favoring the ASD

class in the dataset, which was mostly based on

collections about autism, [10]. There were 367

variables in the 95,577 children's data, 256 of which

were deemed sufficient, and the accuracy was

87.1% in SVM, RFC, and NB. The innovative

intelligence-sharing structure was designed to hide

responsive and simultaneously unstable individual

data. The study also recommended linear variance

analysis as a simple way for keyword separation

(LVA), [11]. SVM was the technique applied in this

filtration procedure. The study's findings offer hope

for employing text mining to safeguard private

health information transmitted over the Internet,

[12]. ASD is predicted using parameters based on

brain activity, 95.9% accuracy was achieved using

SVM with 2 groups and 19 features. The amount of

data was relatively small. It used a cross-validation

approach to extract 6 personality variables from the

data of 851 individuals to train and evaluate their

ML models. This was utilized to categorise patients

into those with and those without ASD, [13]. Facial

expressions from photo to identify psychological

stress is not possible. Using haar cascade algorithm

is used to evaluate the stress via logarithmic

regression, [14], [15].

3 Working Principle

Data processing converts raw data into a more

comprehensible format from a preliminary step.

Figure 1 illustrates the workflow of our proposed

system. The dataset will be preprocessed for missing

data, duplicate data, and noisy data. Pre-processing

of the data can be done by dimensionality reduction

technique. Autism can be predicted using

classification approaches after the dataset is

preprocessed. The efficiency of each classification

may contrasted. The training efficiency will be

greater than the test efficiency if the classification

performs properly. Then, this classification

algorithm may be used for the next training and

classification as the best model.

3.1 Dataset Pre-processing

One of the most essential components in Machine

Learning is the dataset. Once the dataset is

collected, researchers should employ various dataset

types that may depend on their prediction model.

Preprocessing can convert the data into a pattern

that is more easily and productively refined in data

mining, and machine learning through

dimensionality reduction techniques using selection

and extraction.

3.2 Feature Selection

It is the process of contracting the proposal to your

model by using only proper data so the noise in data

can be cleared. The most essential task for building

an efficient classification system. There are three

methods of feature selection are available.

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2023.22.42

Sindhu Veeramani, S. M. Ramesh, B. Gomathy

E-ISSN: 2224-2872

375

Volume 22, 2023

3.2.1 Filter Method

Features are processed based on universal aspects of

the dataset such as interaction with dependent

variables. A faster and better approach when the

sum of features is enormous.

3.2.2 Wrapper Methods

It is the process of removing the weakest feature and

applying a model until a desired feature is selected.

3.2.3 Embedded Methods

It is the process of selecting a feature based on its

weight. It combines the qualities of filter and

wrapper methods.

3.3 Classification Algorithm

It utilizes proposed training data to predict the

expectation that the data will fall into one of the

fixed grades. The training set and testing set were

separated from the entire dataset. 80% of the data

used to train the classification algorithm will come

from the training set, and 20% will come from the

testing set. This sample data partitioning into

training sets and testing sets enables us to evaluate

the overfitting or underfitting of our model. Later

the collection of preprocessed data, the grade

models are enforced to detect ASD. Each

classification algorithm’s performance is assessed

by the level of accuracy it achieves, including NB,

SVM, LR, RFC and KNN

Fig. 1: Workflow of Proposed System

3.3.1 Decision Tree (DT)

It is the most influential and prominent tool for

allocation and prognosis. To forecast the output

class, DT builds the supervised learning model

using a collection of IF-THEN rules from the

training set. Based on characteristics in the dataset, a

hierarchical tree is built. A root node, an internal

node, and a leaf node are the three different types of

nodes that make up a decision tree, [16].

3.3.2 K Nearest Neighbors (KNN)

In this strategy, we compare the new sample with

the training dataset to identify the k examples that

are close to the new sample in terms of class. We

refer to them as neighbors. The classification of this

sample of data is then established via a majority

vote among neighbors, [17].

3.3.3 Support Vector Machine (SVM)

Regression or classification problems are used in

SVM. Every data is plotted as a spot in n-

dimension using the support vector machine, where

n = number of functionalities that possess and each

functionality is the value of a certain coordinate,

[18].

3.3.4 Random Forest Classifier (RFC)

It is a classification algorithm, which has many

decisions trees. Consists of a large number of

personal decision trees that operate as an ensemble.

Each personal tree in the forest spits out a class

prediction. Both continuous variables, as in

regression, and predictor variables, as in

classification, can be included in data sets that RFC

can handle. It produces better results in terms of

classification problems

3.3.5 Naive Bayes (NB)

It makes the process easier to build machine

learning models that can produce accurate

predictions. Learners and classifiers are immensely

quick, when related to refined forms. This helps to

ease issues from the curse of dimensionality, [19].

3.3.6 Logistic Regression (LR)

Logistic Regression is used when the dependent

variable (target) is categorical. It is equal to the

Linear Regression except for how they are used. It

can be used to classify the perception using various

types of data and it can easily resolve the most

efficient variables used for the classification, [20].

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2023.22.42

Sindhu Veeramani, S. M. Ramesh, B. Gomathy

E-ISSN: 2224-2872

376

Volume 22, 2023

4 Evaluation

Evaluation is based on the performance of

classification how much the correctness of the result

of the training system and the testing system. The

scale are listed below:

Table 1. Performance of Classification algorithm

True Positive (TP): The amount of outcomes to be

true in both predicted and diagnosed cases

True Negative (TN): The amount of outcomes to be

false in both predicted and diagnose cases.

False Positive (FP): The amount of outcomes that

are predicted to be false but are really true.

False Negative (FN): The amount of outcomes that

are predicted to be true but are really false.

Accuracy: It is a very important metric in Machine

Learning. It is calculated as the percentage of total

right predictions overall predictions. Figure 2

represent the evaluation graph of accuracy.

Accuracy = 𝐓𝐏+ 𝐓𝐍

𝐅𝐍+𝐓𝐏+𝐓𝐍+𝐅𝐏

Specificity: Another name for specificity is the True

Negative Rate (TNR). It is a capability to estimate a

true negative for each accessible category. Figure 3

represent the evaluation graph of specificity.

Specificity (TNR) = 𝐓𝐍

𝐓𝐍+𝐅𝐏

Sensitivity: Other names for sensitivity include the

True Positive Rate (TPR) or recall. Since it allows us

to see how many situations the model was able to

correctly identify, it is used to evaluate model

performance. Figure 4 represent the evaluation graph

of sensitivity.

Sensitivity (TPR) = 𝐓𝐏

𝐓𝐏+𝐅𝐍

Based on the analysis of classification the

following Table 1. Indication of the matrices of

accuracy, specificity, and sensitivity in various

classification algorithms.

Fig. 2: Evaluation graph of Accuracy

Fig. 3: Evaluation graph of Specificity

Classification algorithm

/ Matrices

Accuracy %

Specificity %

Sensitivity %

SVM

KNN

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2023.22.42

Sindhu Veeramani, S. M. Ramesh, B. Gomathy

E-ISSN: 2224-2872

377

Volume 22, 2023

Fig. 4: Evaluation graph of Sensitivity

From this graph analysis, we can conclude

Random forest classifier and Logistic Regression

have more accuracy when compared with another

classification algorithm.

5 Conclusion

This analysis of the study used several ML

techniques to identify autism spectrum disorder.

Using several performance evaluation measures,

including specificity, accuracy, and sensitivity for

detecting ASD. Having greater accuracy rates of 91%

and 90%, respectively, were the classification

algorithms Logistic Regression(LR) and Random

Forest(RF) Classifier. Future autism spectrum

disorder diagnoses will be more sensitive, specific,

and accurate thanks to a range of machine-learning

techniques.

References:

[1] Guangyao Shen, Jia Jia, Liqiang Nie, Fuli

Feng, Cunjun Zhang, Tianrui Hu, "Depression

Detection via Harvesting Social Media: A

Multimodal Dictionary Learning Solution",

international Joint Conferences on Artificial

Intelligence vol.4, pp. 3838-3844, 2017.

[2] N. Rajaraman and A. P. R. Bhuja, "G

Depression Detection of Tweets and A

Comparative Test", International Research

Journal of Engineering and Technology

(IRJET), vol. 09, no. 03, March 2020, ISSN:

2278-0181.

[3] L. Squarcina, F. M. Villa, M. Nobile, E. Grisan

and P. Brambilla, "Deep learning for the

prediction of treatment response in

depression", Journal of Affective Disorders,

vol. 281, pp. 618-622, 2021.

[4] T. Zhang, A. M. Schoene, S. Ji and S.

Ananiadou, "Natural language processing

applied to mental illness detection: a narrative

review", NPJ digital medicine, vol. 5, no. 1,

pp. 1-13, 2022.

[5] Thabtah, Fadi. An Accessible and Efficient

Autism Screening Method for Behavioural

Data and Predictive Analyses. Health

informatics, 25, 1739-1755 (2018).

[6] Vaishali, R and Sasikala, R, A Machine

Learning Based Approach to Classify Autism

with Optimum Behaviour Sets, International

Journal of Engineering & Technology. 7

(2018).

[7] Sen B, Borle NC, Greiner R, Brown MR. A

General Prediction Model for the Detection of

ADHD and Autism Using Structural and

Functional MRI. PLoS ONE. 13: e0194856.

(2018).

[8] Thabtah, Fadi and Peebles, David.: A New

Machine Learning Model Based on İnduction

of Rules for Autism Detection, Health

Informatics Journal. 26(1) 264-286, (2020).

[9] J. Chen, G. Wang, K. Zhang, G. Wang, and L.

Liu,: A Pilotstudy on Evaluating Children with

Autism Spectrum Disorder, Journal Of

Healthcare Engineering Using Computer

Games. Computers in Human Behavior.

Vol.90, pp.204–214, (2019).

[10] Duda M, Ma R, Haber N, Wall DP.: Use of

Machine Learning for Behavioral Distinction

of Autism and ADHD, Transl Psychiatry. 6:

e732. Doi. 10.1038/tp.2015.221, (2016).

[11] Suman Raj and Sarfaraz Masood,: Analysis

and Detection of Autism Spectrum Disorder

Using Machine Learning Techniques, Procedia

Computer Science, Elsevier, vol.167 pp.994-

1004 (2020).

[12] Vakadkar K, Purkayastha D, Krishnan D.

Detection of Autism Spectrum Disorder in

Children Using Machine Learning Techniques.

SN Comput Sci. 2021; 2(5):386. doi:

10.1007/s42979-021-00776-5.

[13] Deshpande G, Libero LE, Sreenivasan KR,

Deshpande HD, Kana RK.: Identifcation of

Neural Connectivity Signatures of Autism

using Machine Learning, Frontiers in Human

Neuroscience, vol. 7, doi.

10.3389/fnhum.2013.00670, (2013).

[14] Andrew Yates, Arman Cohan and Nazli

Goharian, "Depression and selfharm risk

assessment in online forums", arXiv preprint,

2017, doi.10.48550/arXiv.1709.01848.

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2023.22.42

Sindhu Veeramani, S. M. Ramesh, B. Gomathy

E-ISSN: 2224-2872

378

Volume 22, 2023

[15] Minsu Park, David W McDonald and

Meeyoung Cha, Perception differences

between the depressed and non-depressed users

in twitter, Seventh International AAAI

Conference on Weblogs and Social Media,

2013, 7(1), pp.476-485,

doi.10.1609/icwsm.v7i1.14425

[16] I. Salehin, I. M. Talha, N. Nessa Moon, M.

Saifuzzaman, F. N. Nur and M. Akter,

"Predicting the Depression Level of Excessive

Use of Mobile Phone: Decision Tree and

Linear Regression Algorithm," 2020 IEEE

International Conference on Sustainable

Engineering and Creative Computing

(ICSECC), Indonesia, 2020, pp. 113-118, doi:

10.1109/ICSECC51444.2020.9557394.

[17] M. R. Islam, A. R. M. Kamal, N. Sultana, R.

Islam, M. A. Moni and A. ulhaq, "Detecting

Depression Using K-Nearest Neighbors (KNN)

Classification Technique," 2018 International

Conference on Computer, Communication,

Chemical, Material and Electronic

Engineering (IC4ME2), Rajshahi, Bangladesh,

2018, pp. 1-4, doi:

10.1109/IC4ME2.2018.8465641.

[18] A. Ahmed, R. Sultana, M. T. R. Ullas, M.

Begom, M. M. I. Rahi and M. A. Alam, "A

Machine Learning Approach to detect

Depression and Anxiety using Supervised

Learning," 2020 IEEE Asia-Pacific Conference

on Computer Science and Data Engineering

(CSDE), Gold Coast, Australia, 2020, pp. 1-6,

doi: 10.1109/CSDE50874.2020.9411642.

[19] Islam MR, Sohan M, Daria S, “Evaluation of

inflammatory cytokines in drug-naïve major

depressive disorder: A systematic review and

meta-analysis”2023 International Journal of

Immunopathology and Pharmacology, vol. 37,

pp. 1- 15.

[20] S. Jayawardena, J. Epps and E. Ambikairajah,

"Ordinal Logistic Regression With Partial

Proportional Odds for Depression Prediction,"

in IEEE Transactions on Affective Computing,

vol. 14, no. 1, pp. 563-577, 1 Jan.-March 2023,

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

The authors equally contributed in the present

research, at all stages from the formulation of the

problem to the final findings and solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

This research received no specific grant from any

Funding agency in the public, commercial, or not for-

profit sectors.

Conflict of Interest

The authors have no conflicts of interest to declare.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2023.22.42

Sindhu Veeramani, S. M. Ramesh, B. Gomathy

E-ISSN: 2224-2872

379

Volume 22, 2023