A SHAP-Guided Ensemble Learning Framework for Explainable Autism Spectrum Disorder Identification

Autism spectrum disorder (ASD) is a neurological disorder marked by difficulties in social communication and the presence of limited, repetitive activities. Early detection and accurate diagnosis can significantly improve an individual's symptoms and enhance their daily functioning. However, traditional methods are time-consuming, costly, and often subjective. The aim of this study is to improve the prediction of ASD through the development of machine learning models that utilize algorithms including Random Forest (RF), k-Nearest Neighbors (kNN), Support Vector Machine (SVM), Gradient Boosting (GB), Extra Trees (ET), Multi-Layer Perceptron (MLP), and a stacking ensemble method. We combined ASD screening datasets covering individuals across childhood, adolescence, and adulthood to construct a more comprehensive and diversified dataset. Data preprocessing techniques were applied to enhance data quality. We implemented multiple classification techniques along with a hyperparameter tuning approach. The SVM, MLP and stacking classifier provided the best results among all classifiers. Furthermore, we applied an explainable AI (XAI) approach to interpret the model's predictions, providing insight into feature importance. This study proposes a robust, accurate model that can be used in early detection and aims to support clinicians in their decision-making processes.