AUTISM SPECTRUM DISORDER EARLY DETECTION THROUGH DEEP LEARNING TECHNIQUES

Abstract

Autism Spectrum Disorder (ASD) is a multifaceted neurodevelopmental condition characterized by challenges in communication, behavior, and social interaction. It typically manifests during early childhood and frequently continues into adulthood. Early identification of ASD plays a critical role in enabling timely intervention, which can substantially enhance developmental progress and overall quality of life. This study presents an automated early ASD detection framework based on deep learning, utilizing behavioral responses and demographic characteristics. The AUTISM dataset obtained from the UCI Repository serves as the foundation for model development. Comprehensive data preprocessing was performed, including missing value treatment, categorical feature encoding through label encoding, and normalization of numerical attributes to ensure consistency and improve classification performance. A range of traditional machine learning models—Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Naïve Bayes, and Logistic Regression—were evaluated alongside deep learning architectures such as Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNNs). Although conventional classifiers demonstrated reasonable predictive capability, the CNN model significantly outperformed all others, achieving perfect accuracy on the processed dataset. This application enables users to input screening information and obtain real-time ASD predictions. The proposed system offers an accessible, efficient, and cost-effective solution for preliminary ASD screening, providing valuable support to healthcare professionals, caregivers, and families—especially in underserved regions with limited access to diagnostic facilities.