•  
  •  
 

Abstract

Neurodevelopmental disorders like autism spectrum disorder (ASD) cause significant cognitive, linguistic, object identification, communication, and social skills deficits. Although there is currently no cure for autism spectrum disorder (ASD), early detection can aid in diagnosis and implementing effective preventative measures. Artificial intelligence (AI) tools allow for an earlier diagnosis of ASD than was previously possible. Furthermore, many clinical and not clinical attributes can be used for identification of ASD but select the most proper ones still challenge. Therefore, in this study we propose a Computer-Aided Identification System based on machine learning concept and feature selection methods to diagnosis Children Autism Spectrum Disorder (C-ASD) cases. Two main feature selection methods namely Gain Ratio (GR) and Chi-squared (χ2) that used to rank then select best subset C-ASD attributes. In order to identify ASD cases, four machine learning algorithms are used into proposed system for identification purpose namely, Neural Network (NN), Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM). A dataset contained 1045 cases with 18 attributes employed for feature selection and C-ASD identification processes. C-ASDs are identified into binary classification approach namely negative and positive C-ASD classes. The results indicate only 10 attributes instead of 18 are significant into identification process. We found that each of RF and NN performed better than other classifiers where both classifiers score accuracy reach to 100 % based selected subset. This supports the idea that these models might be used for screening for C-ASD at an early stage of test-bed applications.

Download

Share

COinS