Performance Analysis of TabNet for Autism Spectrum Disorder Classification in Children

Le résumé fourni par la source

Children are the most promising segment of society, but birth disorders can significantly impact a child’s growth and societal presence. Among these, Autism Spectrum Disorder (ASD) is a prominent neurodevelopmental condition characterized by difficulties in social interaction, communication, and repetitive behaviors. Its early detection and proper support can greatly improve the mental conditions of children. There are various detection methods, including eye movement analysis, facial features, questionnaires, etc. Among these, Q-Chat-10 and AQ-10 are widely used for effective screening and assessment. Manual detection of ASD is prone to human-biasness, delayed results, and subject to error based on the stress level of the doctor. As a result, Machine Learning (ML) came into play, but was limited due to manual feature engineering techniques and the inability to deduce hidden relationships. To address it, a Tabular Network (TabNet)-based ASD detection model for children has been proposed that is trained on the autism screening dataset. The model was equipped with Adam optimizer, LRScheduler, early stopping, etc., to ensure proper training. Various experiments were also conducted to gain insight into the dataset and the performance of the classifier that were found to be missing in various studies. An ideal accuracy for multiple split ratios was achieved that demonstrates suitable performance across both balanced and imbalanced datasets. It demonstrated faster training times, enhanced interpretability, efficient feature selection through sparse attention and the ability to handle mixed data types seamlessly.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.