Deep Transfer Learning and Explainable AI Framework for Autism Spectrum Disorder Detection Across Multiple Datasets

Shtwai Alsubai^*

• Prince Sattam Bin Abdulaziz University, Al-Kharj, Saudi Arabia

This paper presents a transfer learning approach for Autism Spectrum Disorder (ASD) detection using Deep Neural Networks (DNN) across three distinct datasets. It first establishes a baseline by training multiple machine learning and deep learning models on a toddler ASD screening dataset from Saudi Arabia, which is augmented with the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. The DNN architecture, featuring regularization and dropout layers, outperforms other models (i.e., LSTM and Attention LSTM). It then leverages this trained model by transferring learned knowledge to two additional ASD datasets, demonstrating improved performance with limited training data. It analyzes model performance through standard metrics and explainable AI techniques, providing insights into key features for ASD Classification across different populations. Results indicate the efficacy of transfer learning for cross-dataset ASD Classification, suggesting the presence of behavioral indicators despite demographic and data collection differences.