Automated Detection of Autism Spectrum Disorder Using Bio-Inspired Swarm Intelligence Based Feature Selection and Classification Techniques

Abstract

Autism spectrum disorders, or ASDs, are neurological conditions that affect humans. ASDs typically come with sensory issues like sensitivity to touch or soundor odour. Though genetics are the main causes, their  early discovery and treatments are imperative. In recent years, intelligent diagnosis using MLTs (Machine Learning Techniques) have been developed to support conventional clinical methods in the domain of healthcare. Feature selections from healthcare databases consume nondeterministic polynomial timesand are hard tasks where again MLTs have been of great use. AGWOs (Adaptive Grey Wolf Optimizations) were used in this study to determine most significant features and efficient classification strategies in datasets of ASDs. Initially,  pre-processing strategies based on SMOTEs (Synthetic Minority Oversampling Techniques) removed extraneous data from ASD datasets and subsequently AGWOs  repeat this procedure to find smallest features with maximum classifications values for recall and accuracy. Finally, KVSMs (Kernel Support Vector Machines) classify instances of ASDs from the input datasets. The experimental results of suggested method are evaluated for classifying ASDs from datasets instances of Toddlers, Children, Adolescents, and Adults in terms of recalls, precisions, F-measures, and classification errors.