Predicting Autism Spectrum Syndrome Employing Explainable Machine Learning and Deep Learning Approaches

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by difficulties in social communication and interaction, along with restricted interests and repetitive behaviors. Early screening can facilitate timely intervention and appropriate support. This study proposes an autism prediction framework using Machine Learning (ML) and Deep Learning (DL) techniques. A publicly available Kaggle dataset containing demographic information, screening questionnaire responses, and assessment scores is used for model evaluation. Decision Tree, Random Forest, CatBoost, LightGBM, LSTM, and BiLSTM models are investigated using accuracy, precision, and F1-score. Experimental results indicate that BiLSTM achieves the highest predictive accuracy among the evaluated models. Future research will focus on larger, diverse datasets and web-based screening applications to improve accessibility and generalizability. The proposed framework supports preliminary screening and does not replace professional clinical diagnosis.