*Corresponding Author: mohitgangwar@gmail.com 51 DOI: https://doi.org/10.52756/10.52756/ijerr.2023.v31spl.006 Int. J. Exp. Res. Rev., Special Vol. 31: 51- 60 (2023) Analysis of Meta-Heuristic Feature Selection Techniques on classifier performance with specific reference to psychiatric disorder Chandrabhan Singh 1 , Mohit Gangwar 2* and Upendra Kumar 3 1 Computer Science & Engineering, Dr. A.P.J. Abdul Kalam Technical University, Lucknow, UP, India; 2 Department of AI & ML, SIRT, Bhopal, MP, India; 3 Department of Computer Science & Engineering, IET, Lucknow, UP, India E-mail/Orcid Id: CS, chandrabhan98@gmail.com; MG, mohitgangwar@gmail.com, https://orcid.org/0000-0001-5654-2317; UK, upendra.ietlko@gmail.com, https://orcid.org/0000-0003-3792-7945 Introduction In any learning technique, feature selection plays an important role. Any learning system can be trained using all its features. Data processed in a learning system consists of many dimensions. Learning any system with all its features will take a lot of computational time and cost. This computational overhead plays a role in introducing intelligent approaches for feature selection. A training system with fewer features, i.e., features with high weightage, will reduce the computational overhead and help in efficient prediction. Classification is an important approach in any learning system. Artificial neural networks (ANN), due to their high accuracy and performance, are the most preferred approach for classification (Schmidhuber, 2015; Galeshchuk, 2016). Due to a lack of explanation, a neural network is combined with fuzzy logic, giving birth to the Neuro-fuzzy concept. The concepts of Neural networks and fuzzy logic are combined to meet human-like decision capability and learning ability. Learning with this type of system provides the optimal, or, say, most reliable, result. Feature selection is the process of finding the subset of the most reliable features from the overall set of features. Selecting optimal features will not only avoid the curve of dimensionality but also simplify the model by reducing its training cost and time (Gangwar et al., 2012; Singh et al., 2020). It will enhance the generalization of the model by reducing overfitting. Figure 1 mentioned below, shows the working of the learning system. Here after normalization, feature Article History: Received: 24 th Apr., 2023 Accepted: 20 th Jun., 2023 Published: 30 th Jul., 2023 Abstract: Optimization plays an important role in solving complex computational problems. Meta-Heuristic approaches work as an optimization technique. In any search space, these approaches play an excellent role in local as well as global search. Nature- inspired approaches, especially population-based ones, play a role in solving the problem. In the past decade, many nature-inspired population-based methods have been explored by researchers to facilitate computational intelligence. These methods are based on insects, birds, animals, sea creatures, etc. This research focuses on the use of Meta-Heuristic methods for the feature selection. A better optimization approach must be introduced to reduce the computational load, depending on the problem size and complexity. The correct feature set must be chosen for the diagnostic system to operate effectively. Here, population-based Meta-Heuristic optimization strategies have been used to pick the features. By choosing the best feature set, the Butterfly Optimization Algorithm (BOA) with the Enhanced Lion Optimization Algorithm (ELOA) approach would reduce classifier overhead. The results clearly demonstrate that the combined strategy has higher performance outcomes when compared to other optimization strategies. Keywords: Feature selection, Meta- Heuristic techniques, optimization, classification