Improved autistic spectrum disorder estimation using Cfs subset with greedy stepwise feature selection technique

The aspiration of this study is to predict autistic spectrum disorder (ASD) using machine learning algorithms. ASD is a weakness in the progress of central nervous system. The reason of ASD can be genetic and/or neurological. The early prediction and detection of ASD can help medical practitioner and family members in better treatment. With the recent development in machine learning, the machine learning algorithms can be used for estimation of ASD. The machine learning algorithms improves the prediction accuracy and also makes the process less time consuming. Machine learning algorithm with feature reduction technique is used for prediction and detection of ASD. The paper presents estimation of ASD using Cfs subset selection with greedy stepwise feature selection technique known as Cfs-GS technique. The Cfs-GS is used for attribute/feature selection. The result of the proposed algorithm has been verified on five different machine learning algorithms with three data sets of different age groups. The results are verified in terms of accuracy, sensitivity and specificity. The result shows that 100% accuracy, selectivity and specificity for child data set can be obtained with proposed feature selector filter for Naive Bayes and Stochastic Gradient Descent classifier. An improvement of 1–36 to 18.27% in accuracy was marked with different classifiers for adolescent dataset. 100% sensitivity and specificity was observed for Random Tree and K-Star classifier respectively for adolescent dataset. For some classifiers the accuracy and selectivity approaches to approximately 100% for adult datasets. The comparative results with other feature selector shows that better accuracy, sensitivity and specificity can achieved by using proposed feature selection technique. The number of reduced features also makes the test less sluggish. So machine learning algorithm with feature reduction technique make the ASD prediction less time consuming along with better accuracy and sensitivity. Moreover, Classifier’s performance can be improved by selecting the accurate feature(s) from the dataset. © 2019, Bharati Vidyapeeth's Institute of Computer Applications and Management.