Aberrant behavior prediction and severity analysis for autistic child through deep transfer learning to avoid adverse drug effect

Abstract

Autism Spectrum Disorder (ASD), in child is identified through various parameters such as social skills, repetitive behaviors, speech and nonverbal communication. Among the above parameters repetitive behavior plays a vital role for physician to prescribe dosage of drugs. The repetitive behavior and more aggressiveness in the autistic child is the symptom for growth of the disease. To control the repetitive behavior, the physician prescribe the dosage level of drug based on Aberrant Behavior Checklist (ABC). The ABC is measured only for few seconds by the physician and such measurement need continuous monitoring for proper prescription of drugs and also to avoid adverse drug effect. The above problem solve through IP Webcam app based ASD recognition for continuous monitoring and replaces the empirical method of ABC measurement. In this paper, the proposed method recognizes behavior and changes in autistic child through activity detection and repetitive behaviour, due to overdosage of drugs. In proposed method, hybrid framework incorporates training of deep CNN model for the monitoring of ASD children in natural environment through Autismdata.Net. Moreover, Transfer learning avoids the over-fitting problem in small size Autismdata.Net dataset through CNN in severity analysis of child. The behavior of ASD children is evaluated through Autismdata.Net dataset and validated through drug thermo regulation of autistic child. Action recognition accuracy of the proposed method is much better than the clinical literate/therapist analysis/observation. The proposed system helps physician for regulation of dosage level to ASD children.