AI-based Early Detection of Autism Spectrum Disorder

Abstract

The diagnosis of Autism Spectrum Disorder (ASD) is generally late due to the use of expert evaluation and also reoccurring observation. To create an early screen for autism, this paper introduces a web platform that will predict ASD from the movement analysis and feature input. There are two types of input available in this system: video upload and CSV data with body movements and gait data. Depending on the type of input used, the model will evaluate the behavior patterns and make predictions with the confidence value. To improve prediction efficiency, the proposed approach employs an ensemble learning technique that integrates three machine learning models: Random Forest, Linear Support Vector Machine (SVM), and a Multilayer Perceptron (MLP) with Principal Component Analysis (PCA). In addition to prediction, the system incorporates video analytics, including frame count, eye contact ratio, and movement score, to provide better insight into the generated results. Experimental evaluation demonstrated excellent performance, achieving an accuracy of 99.31%, with only a single misclassification observed in the confusion matrix.