Usage of ML Techniques for ASD Detection

Abstract

Autism spectrum disorder (ASD) is one of the severe neurodevelopmental illnesses across the world which influences the communication skill and behavior of a person. Early detection and identification of ASD can significantly minimize the effect. A mobile-based ASDTest app with a set of screening questionnaire was instigated as a solution against the lengthy and expensive diagnosis procedures for ASD. The app collected information from all categories of individuals with more than 1400 cases which were stored in Kaggle as well as UCI machine learning (ML) repository to carry out research. The chapter identifies the ASD class in the following steps: The investigation is incorporated with mean standard deviation method for standardization followed by applying singular value dimension (SVD), large margin nearest neighbor (LMNN), and t-distributed stochastic neighbor embedding (t-SNE) for feature dimension reduction and further classified the ASD class type using various classifiers such as surface vector machine (SVM), Naïve Bayes (NB), decision tree (DT), and k-nearest neighbor (KNN) ML classifiers. The performance of classifiers is investigated by utilizing different evaluation parameters. The investigation is carried out upon the data mentioned in the former datasets. The result indicates that, dimension reduction in addition with ML classifiers yield clinically acceptable outcomes for effective identification of ASD.