Efficient Feature Selection using Particle Swarm Optimization: A hybrid filters-wrapper Approach

In machine learning, feature selection can be used to reduce the computation time and improve the learning accuracy, especially when dealing with high-dimensional data sets. Particle Swarm Optimization (PSO) has attracted significant concerns to enhance the feature selection process due to its efficiency in solving problems. This paper introduces a new hybrid filters-wrapper approach that is used to enhance the feature selection process using PSO algorithm. Our proposed approach combines five filtration methods in with different weights to produce a new hybrid filters-wrapper algorithm using BPSO. The proposed approach has been evaluated by performing comparisons with other methods like wrapper alone and filter alone. The obtained results show that the proposed approach has achieved better performance than other approaches taking into account three parameters; The number of selected features, the classification accuracy, and the execution time. In addition, the new approach has been tested to ensure its stability in the feature selection and it has shown a high degree of stability.