Probe mechanism based particle swarm optimization for feature selection.

Feature Selection (FS) is regarded as an important preprocessing technique with the main aim of discarding irrelevant and redundant features on the premise of improving or retaining classification accuracy. As one of the most popular FS methods, the swarm intelligent (SI) algorithm based FS method faces great challenges in solution quality. Given this, a novel probe mechanism based particle swarm optimization (PPSO) is proposed. The probe mechanism is proposed to test the dataset with different filter methods by employing a part of features to generate high-quality initial population, thus playing a critical role in enhancing the final results of the algorithm. Next, the repeated binarized solution avoid mechanism prevents the same binarized solution from repeating compute the fitness function, which can improve the exploration capability. Additionally, a volcanic eruption mechanism based on the updated information of the global best solution is designed to improve the performance in convergence rate and search precision. In the experiment section, 14 UCI datasets are employed to assess the performance of the PPSO. The outcomes show that the designed strategies can significantly improve the original PSO, and the PPSO is more efficient than the other comparative algorithms for solving the FS problems. In addition, the impact of each strategy is analyzed, and the volcanic eruption mechanism is the most efficient among the three strategies. [ABSTRACT FROM AUTHOR]