Design of Diagnostic Framework for Detecting Autism Spectrum Disorder using Conditional Mutual Information Maximization-Random Forest (CMIM-RF) Approach

Abstract

Autism Spectrum Disorder (ASD) has detrimental effects on social interaction, communication, and repetitive behaviors. Standardized diagnosis approaches for ASD often include extensive, subjective clinical observations and behaviour assessments. Due to recent advances in Machine Learning (ML), more precise and time-efficient ASD diagnosis may soon be achievable. Using machine learning approaches such as data pre-processing, feature extraction, and classification, this study proposes a diagnostic paradigm for ASD. In this work, it was analysed that the most advanced machine learning algorithms used in each component and compared their performance on a benchmark dataset. In terms of accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve, proposed Conditional Mutual Information Maximization-Random Forest (CMIM—RF) method was found to be superior to conventional diagnostic methods. According to the findings of this study, ML-based diagnostic frameworks may one day become an indispensable resource for clinicians and researchers in the area of ASD. However, more study is required to validate these findings in larger samples and evaluate the potential for bias. Physiological data, genetic information, and brain imaging are examples of additional aspects that might be used in an attempt to improve diagnosis accuracy.