LAUSR

Neighborhood rough set is an excellent mathematical tool to carry out feature selection on both numerical and categorical dataset. However, conventional feature selection algorithms usually use greedy heuristic search strategies, which is easy to trap in the local extreme point. In this study, we propose a hybrid feature selection model that combines neighborhood rough set with an improved particle swarm optimization. The model computes the dependency degree of decision from neighborhood rough set as the objective function to evaluate the selected features, and then takes advantages of PSO’s stochastic search to discover the optimal solution more effectively. In order to improve the global search ability and alleviate the stagnation in local optima, the model modifies PSO part by adopting genetic operators and Levy flight. To evaluate the performance of this model, we implement experiments using twelve benchmark datasets and two classifiers (kNN and SVM). Compared with five representative filter-based approaches, experimental results show that our model can not only guarantee the stronger classification ability but also remove more redundant features in most datasets.