Early diagnosis of Autism using Machine Learning techniques and Gated Recurrent Units

Detecting Autism Spectrum Disorder (ASD) has always been a challenging task. Several behavioral changes are reflected in an ASD's patient, which can be targeted by modern day technologies like Machine and Deep Learning (ML/DL) to improve the detection process. This can help in an efficient early diagnosis, and boost the entire process for doctors, beginning from conducting any tests or detailed specific diagnosis, to implementing appropriate measures for treatment. The objective of this paper is to train and validate various models (Decision Tree, Random Forest, SVM, KNN and Gated Recurrent Units or GRUs), and conduct a comparative analysis based on various performance metrics. We were able to achieve a 100% training and validation accuracy with both, Random Forest and GRUs, after rigorous tuning of the models. KNN classifier yielded the poorest results.