OPTIMIZED HYBRID LEARNING APPROACH FOR AUTISM SPECTRUM DISORDER DETECTION

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition marked by difficulties in social interaction, communication, and repetitive behaviours, making early and accurate detection particularly crucial in resource-constrained regions. To address this need, this study introduces a hybrid machine learning approach that leverages the Grey Wolf Optimizer (GWO) for hyperparameter tuning and XGBoost for enhanced classification, improving ASD detection using facial image data from a publicly available kaggle dataset. The dataset comprises images of individuals with and without autism across various age ranges, providing a diverse foundation for analysis. The study evaluates two deep learning architectures—VGG16 and MobileNet—optimized through GWO to fine-tune critical parameters such as batch size, learning rate, and optimizer selection. Additionally, XGBoost is employed to ensemble weak classifiers, enhancing model robustness when dealing with imbalanced datasets. The framework is designed for automated image processing and includes a comparative analysis of classifier performance.

Experimental results highlight the effectiveness of the GWO-optimized VGG16 model, which, when configured with a batch size of 2 and the Adam optimizer, achieves 99% validation accuracy and 87% testing accuracy, along with an F1-score of 88%, precision of 85%, and recall of 90%. Further validation on an independent ASD dataset confirms the model's real-world applicability, maintaining 85% accuracy. Key contributions of this research include the use of GWO-driven optimization to refine hyperparameter selection in deep learning models, boosting-enhanced classification to mitigate dataset imbalances and improve diagnostic precision, and the development of a cost-effective, scalable solution for ASD detection in low-resource settings through facial image analysis. By combining optimized deep learning with ensemble techniques, this study demonstrates the potential for accurate, accessible, and automated ASD diagnostics, offering a promising tool for early intervention in underserved communities.