Development of Diagnostic Model for Early Detection of Autism Spectrum Disorder in Young Children Using Transfer Learning Based AlexNet

In recent years, Autism Spectrum Disorder (ASD) is neurological disorder affecting communication then behavior. Symptoms such as social difficulties and language delays. Accurate diagnostic tools and increased awareness are essential for effective support and inclusion. To overcome these issues proposed an approach for early ASD detection using a Transfer Learning (TL) based AlexNet model. Leveraging the ASD dataset, Quantitative Checklist for Autism in Toddlers-10 (Q-Chart-10) used for preprocessing to distinguish between autistic and non-autistic patients. Correlation-based Feature Selection Harmony Search (CFS-HS) is then applied to identify the most relevant features. A modified AlexNet classifier, built upon pre-trained models, is trained and validated using the classified data. This approach enables the effective reuse of pre-trained models, leading to accurate early-stage autism detection offering a promising tool for opportune intrusions and enhanced consequences. The proposed model has the possible to develop early ASD detection, allowing healthcare professionals to provide targeted support and interventions, ultimately leading with better outcomes for individuals with ASD. The proposed model obtained high results, in the context of obtained high results in the context of accuracy with 98.97%, Area Under the Curve (AUC) with 98.99%, specificity with 98.92% when compared to existing methods like Support Vector Machine (SVM).