Chaotic Ant Swarm based Feature Subset Selection with Concept Drift Detection and Classification Model for Data Streaming Applications

Abstract

Recently, high dimensional data stream poses additional challenges to statistical classification process owing to the varying distribution of the training and target data over time, which is termed concept drift. The concept drift explores the considerable modifications in data tuple which has the significance of raising the contextual problems in the expected output of the methods depending upon the corresponding data tuples. Though several concept drift detection models are available in the literature, it is required to enhance the drift detection performance. In this view, this study develops a Chaotic Ant Swarm based Feature Subset Selection with Concept Drift Detection and Classification (CASFS-CDDC) technique. The major aim of the CASFS-CDDC technique is to choose an optimal subset of features prior to concept drift and classification processes. The proposed CASFS-CDDC technique involves the design of CASFS technique for the selection of feature subsets. Moreover, early drift detection (EDD) technique is applied for the detection of concept drift. Furthermore, auto encoder (AE) is used for the classification of data into appropriate classes. In order to ensure the enhanced performance of the CASFS-CDDC technique, an extensive experimental analysis is carried out and the comparative study reported the better performance over the existing techniques.