Abstract:
Neurons are the basic building blocks of the human brain and nervous system. In the past few
years, Neurodevelopmental Disorders (NDD) in toddlers has been increasing in the world year
by year and this is becoming a major issue in pediatric portion health. Specifically, Autism
Spectrum Disorder (ASD) and Attention Deficit Hyperactivity Disorder (ADHD), are significantly
affecting complex disorders, for children’s main motor function. Early intervention for
these disorders is strictly effective to improve outcomes of toddlers. However, due to the traditional
screening techniques and lack of clinical resources in Sri Lanka, there is a massive delay
in diagnosis. This research aims to fill this gap by early recognition disorders and providing
more accurate results. This exploratory study proposes a static, Tab-Net based interpretable
deep learning model by analysing behavioral symptoms and signs in children below 36 months
age. Evaluated the model using quantitative strategy applied a merged dataset of behavioral
analytics. Tab-Net model differs from the traditional black box classifications by interpretable
feature selection evaluations. Various models were built using same dataset including Decision
Trees (95.40%), SVM (95.40%) and Random Forest (95.82%) to compare the performance of
the TabNet model. The Tab-Net model achieved impressive performance with 97.07% testing
accuracy, 97.08% precision, 97.07% recall and F1-score, and 99.46% ROC-AUC. The results
demonstrate that Tab-Net provides competitive performance with early NDD diagnosis in resource
limited settings.
