A Review of Machine Learning Models for Predicting Autism Spectrum Disorder

Author Name(s): P. Kanchanamala, G. Leela Sagar
Author Email: kanchanamala.p@gmrit.org

Abstract

Autism spectrum disorder (ASD) is a neurological and developmental disorder that impacts the behavior of the person throughout a life. Every individual with ASD exhibits the difficulty in communication and social interaction with restricted interests and repetitive behaviors. There is no standard diagnosis and treatment for ASD. The social behavior of the children can be improved by early identification of autism spectrum disorder (ASD). A vigorous predictive mechanism desires features such as facial expressions, eye movement, and brain activity images. The research embody diverse approaches to classify and categorize cognitive disorders affected children based on fMRI scan images, facial expressions, gage contingent eye tracking and stereotypical motor movement. Predictive models can be developed to determine if the new patients are developing with ASD using historical patient data. This paper explores the machine learning models built using structured clinical patient data for predicting the ASD subjects

Introduction

This paper aims to review the various approaches considered and the machine learning models built for early prediction of the Autism Spectrum Disorder (ASD). ASD is a brain development disorder that limits the natural growth of the children in terms of communication and social behaviors [1]. The American Psychiatric Association has created a guide, Diagnostic and Statistical Manual of Mental Disorders (DSM-5), to diagnose mental disorders. As per DSM5 people with ASD suffers with restricted interests, repetitive behaviors which limits them with communication and social interaction difficulties [2]. Due to lack of awareness on the disorder parents ignore the early signs that are exhibited by their children. Hence it is very much necessary for the early prediction of the disorder [3]. Early diagnosis of ASD within the first few years of life allows for treatment while the child’s brain is still rapidly developing. Autism spectrum disorder (ASD) diagnoses are on the rise, due in part to physicians adopting more standardized methods for evaluation and diagnosis. Due to lack of standard diagnosing methods and treatment it is difficult to diagnosis ASD. Doctors observe the child’s behavior and development to make a diagnosis which needs a lot of time and effort to detect ASD in an individual. Machine learning is a mathematical learning framework for extracting useful patterns from the available large amounts of data. Extracted knowledge can be helpful to humans in clinical decision making. This computational learning methodologies are being rapidly used in the study of neurocognitive disorders [4].

Conclusion

The Centers for Disease Control and Prevention (CDC) estimates autism’s prevalence as 1 in 59 children which includes 1 in 37 boys and 1 in 151 girls. Early identification of ASD is essential to improve the child’s ability to learn new skills. Manual diagnosis of the ASD is a much time consuming process for the collection as well as identifying the disorder. The use of Machine learning methods reduces the time and man power and increases the accuracy in detecting the ASD from Typically Developed Controls (TC). This study reviews the machine learning models used to classify the ASD subjects and TC using various input parameters viz. Stereotypical Motor movements, Eye Tracking, Brain Images.

1,673 total views, no views today

Download PDF File

About the author: admin