Early Detection of Autism Spectrum Disorder Using Machine Learning

Autism is classified as a developmental disorder and primarily disrupts social interactions and communication. This disorder has no definitive treatment, making early diagnosis crucial for mitigating its effects. The purpose of this study is to identify autistic individuals based on the recorded information of their walking pattern by Kinect sensor.

In this research, the machine learning method was employed to identify autistic individuals based on recorded joint position data during walking, recorded by the Kinect sensor. First, a group of statistical features was extracted from the Kinect data, which included joint positions and the angles between them. Then, the extracted features were evaluated using the statistical test of analysis of variance, and the optimal features were selected. Finally, classification was performed by decision tree classifier.

In this research, the classification of healthy and autistic individuals was done by the decision tree classification and 42 optimal features selected based on statistical analysis, and the accuracy of classification was 85%. The sensitivity and specificity obtained in this classification are 88 and 82%, respectively.

According to the classification results, this research was able to achieve acceptable accuracy by using the low dimension feature vector obtained by statistical analysis. This research, shows autistic individuals can be classified from healthy people only by having the position of several joints. It is suggested researches in future, using this method for measurement the recovery rate or control autism in patient after performing treatment methods.