Bi-Directional Feature Fixation-Based Particle Swarm Optimization for Large-Scale Feature Selection

Feature selection, which aims to improve the classification accuracy and reduce size of selected feature subset, is an important but challenging optimization problem in data mining. Particle swarm (PSO) has shown promising performance tackling selection problems, still faces challenges dealing with large-scale Big Data environment because large search space. Hence, this paper proposes a bi-directional fixation (BDFF) framework for PSO provides novel idea space selection. BDFF uses two opposite directions guide particles adequately subsets different sizes. Based on directions, can fix states some features then focus others when updating particles, thus narrowing Besides, self-adaptive strategy designed help concentrate more direction stages evolution achieve balance between exploration exploitation. Experimental results 12 widely-used public datasets show that obtain smaller higher accuracy.