Early Stage Autism Spectrum Disorder Detection of Adults and Toddlers Using Machine Learning Models

Abstract

Autism Spectrum Disorder (ASD) is a major neurodevelopmental disorder that limits communicative, cognitive, linguistic, and natural skills and abilities. Children or adults with ASD may behave differently from other children or adults, but it is not an illness. Autism is not curable, although early intervention can help lessen symptoms. There are a wide variety of types of ASD and the levels of severity and symptoms. In this study we gathered ASD data of two different stages: one is an adult group and another is toddlers, and we applied seven different classification techniques and assessed their performance. In the toddler dataset, the light GBM, DecisionTree, XGBoost, RandomForest and Gradient Boosting have given us the best performance with a 100% accuracy score. The XGBoost, on the other hand, had the greatest performance in the adult dataset, with a score of 97.58% accuracy. Following these classifications, we evaluated their performance in both datasets using accuracy, recall, and f1 score. We also calculated the significant and associative features of both datasets by using statistical and machine learning methods. We have also identified the features that can be useful to classify Children with ASD. These analytical approaches indicate that machine learning methods can correctly predict ASD status if the dataset is appropriately optimized. It suggests that we can apply these models for ASD detection in the early stages. We believe our approach might be used to detect ASD in its early stages for diagnostic purposes.