A deep learning predictive classifier for autism screening and diagnosis

Abstract Autism spectrum disorder (ASD) is a developmental disorder that affects communication and behavior. An early diagnosis of neurodevelopmental disorders can improve treatment and significantly decrease associated health-care costs, which reveals an urgent need for the development of ASD screening. However, the data used for ASD screening are heterogenous and multisource, resulting in existing screening tools for ASD screening which are expensive, time-intensive, and sometimes fall short in predictive accuracy. In this chapter, we apply novel feature engineering and feature encoding techniques, along with a deep learning classifier for ASD screening. Algorithms were created via a robust deep learning classifier and deep embedding representation for categorical variables to diagnose ASD on the basis of behavioral features and individual characteristics. The proposed algorithm is effective compared with baselines, achieving 99% sensitivity and 99% specificity. The results suggest that deep embedding representation learning is a reliable method for ASD screening.