A Novel Binary Optimization Algorithms for Feature Selection Problem

Abstract

Feature selection (FS) become more important for getting insights from large dimensional dataset. FS, a pre-processing step that select relevant and matched significant newlinefeatures from entire dataset. FS process is classified into 2 type namely; filter based and wrapper based methods. Classical optimization techniques repeatedly fail to construct global optimisation due to the large search area. Recently, several hybrid models incorporating various search policies have been offered; however, mostly these models are deal with low dimensional datasets. In this work, different types of feature selection method with different optimization algorithm is discussed. Initially, wrapper based newlinemethod that employs BGEO-TVFL method is proposed to minimise FS problems. This newlinemethod trace optimal FS based on GEO algorithm. TVFL maintains a stability between newlineboth exploitation and exploration of GEO. Performance of BGEO-TVFL is analysed newlineby comparing with several metrics under various UCI (UC Irvine) datasets. Proposed newlineBGEO-TVFL method obtained best result than compared methods and is right choice newlinefor dealing with dimensionality decrease difficulties. Next, HBFA-GA method is pro- newlineposed to overcome FS problem using a wrapper model. HBFS-GA executes on continuous search, but the FS is in discrete space. By utilizing transfer functions, continuous search has been transformed into a discrete one. To analyze the best transfer function newlineand investigate HBFS-GA, proposed model used eight distinct transfer functions. The newlineproposed HBFS-GA can fit compact dimensionality reduction problems and become a dominant feature subset selection tool. Finally, a CBPOA method was implemented to newlineselect the optimal feature selection. Chaos theory is included in this method to reduce newlinethe slow convergence rate and local optimal issues in traditional POA. In CBPOA, eight newlinetransfer functions are used to find the best one and inspect CBPOA. CEC-2017, 2018 newlineand 2020 benchmarks datasets are utilized to validate the performance and efficiency newlineof these three im