Predicting Autism Spectrum disorder by Machine learning from Healthcare Communities

Abstract

Autism Spectrum Disorder is a syndromic disorder related to neurological elements resulting to the problems in communication, interacting socially, behavioral, and sensory. According to World Health Organization, the count of patients detected with Autism Spectrum Disorder is slowly increasing. Recent studies concentrate on data collection, Clinical analysis, and brain image laboratory analysis. They do not concentrate much on diagnosing Autism based on Artificial Intelligence and Machine learning.

Goal: This paper mainly intends to classify and categorize Autism data to give an understandable, rapid and simple means to help early intervention of Autism Spectrum Disorder.

Methods: Three groups of Autism Spectrum Disorder datasets are taken for Child, adolescence, and adults. We applied k-Nearest Neighboring, Support vector Machine and Random Forest algorithm to classify the Autism Spectrum Disorder data. During our experimentations, the data was split at random into training sets and test sets. The sections of data were picked at random to assess the classification algorithms.

Outcomes: The outcome and results were evaluated by average values. This is proven that Support Vector Machine and Random forest are efficient algorithms for Autism Spectrum Disorder classification. In specific, Random forest algorithm classified with 100% accuracy for all datasets.

Conclusion: It is been observed that early intervention is possible absolutely. The accuracy of diagnosing Autism Spectrum Disorder will be higher if the data samples count is huge. The results show that Support Vector Machine and Random Forest algorithms gives good classification score compared to k-Nearest Neighboring algorithm w.r.t accuracy, F-measure, sensitivity, and Area Under Curve. We found that Random Forest algorithm is efficient and effective compared to Support Vector Machine and k-Nearest Neighbouring algorithm for data classification.