Early Detection of Autistic Children Using Wrapper-Based Feature Selection Technique

Abstract

AbstractSelection of features is a crucial technique in ML classifiers, especially for datasets with a lot of dimensions. Feature selection is a popular machine learning method in which subsets of the data’s available features are chosen for use in a learning algorithm. The remaining, insignificant dimensions are removed from the most excellent feature, which has the fewest number of dimensions that contribute the most to precision. The goal of the selection of features is to choose a subset of info factors by eliminating characteristics that have practically or no prognostic value. Strategies for choosing elements can be partitioned into three categories. Filter strategies are one, Wrapper techniques are another, and Embedded strategies are the third. Our main goal is to develop a subset feature for autism spectrum disorder premature prediction using several wrapper-based feature selection algorithms. Autistic is a group of neuro-developmental disorders characterized by societal communiqué difficulties, restricted interests and activities, and abnormal tactile sensitivities. This study looks into the use of wrapper features selection techniques such as sequential forward selection (SFS), sequential backward selection (SBS), Sequential Backward Floating Selection (SBFS), Sequential Forward Floating Selection (SFFS), and Recursive Feature Elimination (RFE) as well as optimal selection approaches based on classifiers like RF, GBC, and CART. According to this study, the search methodology employing RFE based on the RF algorithm outperformed other methods in terms of average accuracy of 87%. The advantage of selecting feature subsets is that they are more accurate and take less time to run.KeywordsFeature selectionWrapper-based techniqueASD