Deep Analysis of Autism Spectrum Disorder Detection Techniques

Abstract

Communication and social relations and behavioral disabilities come under ASD, a complex neurodevelopmental disorder. Early diagnosis has to be performed to intervene promptly to improve someone's developmental outcome who has such a disorder. This research provides an enhanced predictive framework for an early approach and diagnosis of ASD through state-of-the-art deep and machine learning methods. Two datasets are used to build a strong and robust classification model the Autism Prediction Dataset and the Autism Spectrum Disorder Screening Data for Toddlers in Saudi Arabia. Feature engineering, data preprocessing, and various algorithms are utilized in the methodology to assure high prediction accuracy. In addition to standard machine learning (ML) techniques of Random Forest and SVM, deep learning architectures such as Neural Network, MobileNet, Vision Transformer, MobileNet+LSTM, and GNN over band gradient boosting are experimented with. The main comparative measurements used for evaluating the models are accuracy, precision, recall, and F1-score. Additionally, to improve trust and transparency in clinical use, we can use Explainable AI (XAI) techniques are used to offer insights into model decisions. According to experimental findings, the performance of ASD prediction systems is considerably improved by integrating these cutting-edge models. The suggested strategy has a great chance of bolstering early screening initiatives and helping medical professionals make better-informed diagnostic choices. Real-world implementation and ongoing model improvement using bigger and more varied datasets are potential future directions.