Autism, or Autism Spectrum Disorder (ASD), encompasses a range of neurodevelopmental conditions characterized by variations in behaviors, speaking skills, social relations, attentive power, ability to learn, and interpersonal skills. Although typically diagnosed in childhood, a significant number of adults with autism don’t get diagnosed in their childhood or better understand their symptoms with age. This study focuses on comparing various demographic attributes associated with ASD and demonstrates the application of the TabNet model for autism screening in an adult population. The TabNet model achieves an impressive accuracy rate of 98% and provides feature importance scores- identifying result, A6Score, CountryOfRes, and A9Score as the most significant features from the dataset contributing to autism. To assess its interpretability, a comparison between TabNet and Multi-Layer Perceptron - SHAP concerning global features is conducted, which ranks the results column lower. This comparison illustrates the advantages of employing a non-black-box deep learning technique, highlighting its potential benefits to eXplainable AI in the context of medical diagnosis.
