Hybrid Optimization Algorithms for Enhancing Feature Selection

Abstract

Data mining plays a pivotal role in analyzing extensive datasets to unveil hidden newlinepatterns and relationships. Feature selection, a critical preprocessing step in data mining, improves predictive models by identifying relevant characteristics and discarding irrelevant ones, thereby enhancing classifier accuracy. However, selecting the most appropriate features from large datasets poses a significant challenge. In this research, a novel Chaos Quasi-Oppositional based learning and Simulated Annealing merged with Flamingo Search Algorithm (CQOFSA-SA) is introduced for feature selection. This approach utilizes Flamingo Search Algorithm (FSA) augmented by Simulated Annealing newlineand Chaos Quasi-Oppositional based learning (CQOBL) to identify the optimal feature subset, effectively reducing dataset dimensionality. The CQOFSA-SA approach newlineachieves an average increase of 10% in classification accuracy and reduces dataset dimensionality by an average of 60 features compared to existing methods. newlineAdditionally, a hybrid feature selection strategy, the Quasi-Oppositional based Flamingo Search with Generalized Ring Crossover (QOFSA-GRC) model, is proposed to address the challenge of selecting the most relevant feature subset based on specific criteria. newlineQOFSA-GRC integrates quasi-oppositional learning and Flamingo search in two popu- newlinelations, thereby mitigating the curse of dimensionality. Generalized Ring Crossover is newlineemployed to select important attributes from datasets, with validation conducted using newlinethe Kernel Extreme Learning Machine (KELM). The recommended strategy signifi- newlinecantly outperforms alternative approaches, yielding an average improvement of 15% in newlineclassification accuracy across various datasets. newlineTo address challenges related to dimensionality and feature selection, multi-objective optimization algorithms are employed. The NCOBL-MOFSA multi-objective feature selection strategy, which integrates the Neighbourhood Centroid Opposition-Based Learning Mutation Operator with the Multi-Objective Flamingo Search A