Early Autism Detection Using Machine Learning-Based Behavioral Analysis

Abstract

Timely interventions to detect ASD early in life are important for enhancing developmental outcomes in the long term. Conventional diagnostic methods can be expensive, subjective, and time-consuming, requiring clinical experts and multiple examinations. This study aims to design a multimodal machine learning system to facilitate the early identification of autism using three independent modalities: behavioral questionnaires, facial feature extraction using CNNs, and eye-gaze tracking via a live webcam. Behavioral, visual, and neurobehavioral indicators are analyzed individually by each subsystem, and their results are combined through an ensemble module to produce a final classification by majority vote. The system was evaluated on a small pilot dataset of 10 children (5 autistic, 5 neurotypical). The ensemble model achieved 80 percent accuracy, with a precision, recall, and F1-score of 4/4, 4/4, and 4/4, respectively, for four true positives, four true negatives, one false positive, and one false negative. These results suggest that the proposed approach is feasible for detecting ASD in its early stages. Although this study is limited by a small sample size and controlled conditions, it demonstrates the potential of the approach. AI-based systems can assist in the initial screening of children with ASD, particularly in low-resource settings. Future studies can incorporate larger datasets, concurrent clinical validation, and longitudinal behavioral traits, which may further enhance the system’s predictive performance.