Empirical Evaluation of Machine Learning Techniques for Autism Spectrum Disorder

Autism spectrum disorder (ASD) is a complicated neurological system disorder that has severe effects on a person’s spoken communication, logical thinking, and personto-person interaction. The early symptoms start looking visible at the age of two to four years and the main ground of these symptoms could be hereditary or ecological. Presently, the standardized approach for detecting ASDs is usually very timeconsuming and the cost of medical care is extravagant. Therefore, early detection of ASD would be helpful. Machine Learning (ML) algorithms have the required potential to predict the chances of having autism or not, based on the type of dataset provided to it and the kind of algorithms that are effectively applied to it. The focus of this study is on screening data sets and the use of machine learning algorithms, including XGBoost (XGB) (Ensemble Technique), Random Forest Classifier (RFC), Support Vector Classifier (SVC), Logistic Regression (LR), and Artificial Neural Network (ANN). Our aim for this research was to develop a model that shortens the process of identification of ASDs at an early stage. The accuracy and precision scores of 92.2% and 0.88, respectively, indicate that our model has done well. Our research could be extended in developing a large-scale model that contains a large dataset having a variety of attributes and the integration of brain and facial MRI scans could be helpful in identifying ASD.

Keywords—Artificial neural networks, Autism spectrum disorder, Machine learning, Random forest classifier, XGBoost