Ensemble Transfer Learning Techniques for Classification of Autism Spectrum Disorder

Autism Spectrum Disorder is a complex neurological condition, becomes evident in children during the initial three years of their lives, Significantly influencing social dynamics, communication patterns, and behavioral norms. In this study, we introduce an ensemble learning-based classification model utilizing sMRI data to differentiate between control and Autism patients in the ABIDE dataset. The proposed method for precise categorization of Autism Spectrum Disorder relies significantly on the pivotal role of Convolutional Neural Networks. We utilize three pre-trained models, namely Vggl9, Densenetl21, and Mobilenetv2, which have been fine-tuned through transfer learning to optimize their performance. The evaluation process incorporates ensemble weighted averaging, and grid search optimization is applied to enhance the effectiveness of these models. The synergy of ensemble weighted averaging and grid search optimization proves to be highly effective, resulting in a 97% accuracy, 100% precision, 100% recall, 97% Fl-score, 97% specificity and 96% AUC score when validated using the ABIDE Dataset. This achievement signifies a noteworthy 1 % improvement over the accuracy attained by the best-performing individual model.