Optimizing Feature Selection in Machine Learning with E-BPSO: A Dimensionality Reduction Approach

In the era of informatics, the effectiveness of machine learning models is compromised due to the challenge of dimensionality in the data. The presence of redundant and irrelevant features significantly increases computational complexity, posing a central obstacle in the extraction of valuable insights from the extensive dataset. Any machine learning model’s performance suffers because of the issue of the plague of dimensionality. To improve the classifier’s performance, feature selection is applied beforehand on applying the machine learning model. Feature selection is accomplished using Enhanced Binary Particle Swarm Optimization (E-BPSO) with the aid of boosting the performance of the K-Nearest Neighbor (K-NN) classifier and is experimented on benchmarking real-world datasets. The conventional BPSO suffers from the problem of exploration which leads to premature convergence. In order to overcome the drawbacks of conventional BPSO, E-BPSO is proposed. The enhancement is made by integrating the self-adaptive velocity to drive the particle with the aid to balance exploration and exploitation. The performance of the proposed E-BPSO is evaluated against the traditional binary particle swarm optimization algorithm and genetic algorithm, considering metrics like accuracy, fitness, root mean square error, and dimensionality reduction ratio.