Predictive Analysis of Autism Spectrum Disorder (ASD) using Machine Learning

Autism Spectrum Disorder (ASD) is a severe condition related to brain development that impairs communication and interaction with others. It is a “developmental disorder” because symptoms generally appear in the first two years of life or later. Till that age, children behave pretty similar to typically developing individuals. Tracking the symptoms leading to autism is challenging because of the wide variation in the type and severity of the disorder. One of the pitfalls is that many countries, including Pakistan, do not track or report their autism cases. In this paper, we have shown how we performed predictive analysis after collecting data from the only government-run Child Psychiatry Department in Pakistan, Mayo Hospital, Lahore. We also collected data from a private school made explicitly for children with autism in Lahore. Data of a total of 100 autistic children under the age of 12 was collected. To avoid any class imbalance problem, we also collected data of 100 healthy children. After performing various machine learning algorithms on our data, our model managed to classify children with autism with an accuracy of 95%. To the best of our knowledge, no such work has been done regarding autism in Pakistan before. This research aims to create a supervised machine learning model that can classify ASD individuals in much less time and effort compared to currently practiced diagnostic procedures and help us visualize common causes leading up to it so that appropriate lifestyle changes can be made accordingly. In the future, the number of records can be increased, and accuracy can be maximized even further. Our model can be used by various physicians and parents of children with ASD for early prediction and for visualizing common causes leading up to it.