Feature Selection Methods Based On Grey Wolf Optimization

In many areas such as data mining and machine learning, data sets may contain a large number of features. However, the redundant or irrelevant features may reduce the classification performance. As one of the effective ways to solve this problem, feature selection has become the focus of attention for a long time and a lot of work has been done. With the accumulation of data, new practical problems continue to emerge, so the demand for new feature selection methods is also increasing. The intelligent optimization algorithm has a strong search ability, and it can find the near-optimal solution in a short time. Therefore, it is a hotspot in the field of data mining to solve the feature selection problem by using the intelligent optimization algorithm. Grey wolf optimization (GWO), as a new intelligent optimization method, has a strong global search capability. In this paper, we introduce the GWO algorithm for the feature selection problem in-depth study.Firstly, in order to improve the evolution mechanism of grey wolves, a hybrid algorithm (HGAGWO) is proposed, which incorporates the genetic algorithm (GA) into GWO, and it is used for feature selection.In HGAGWO, N grey wolves are generated randomly, and then the fitness values of them are calculated by the fitness function. The N grey wolves are sorted by descending sequence according to their fitness values. The first half of the N grey wolves are optimized by GWO, then the optimized grey wolves are saved as the first half of the N middle grey wolves and are optimized by GA. The twice-optimized grey wolves are saved as the remainder half of the N middle grey wolves. As a result, the N middle grey wolves are generated. Keep the iteration ceaseless until the HGAGWO algorithm stops.Secondly, in order to improve the population initialization mechanism of grey wolves, an improved grey wolf optimization algorithm (IGWO) based on GA is proposed and used for feature selection. In IGWO, at first the GA is used to generate the diversified initial positions of the grey wolf population, and then the GWO algorithm is utilized to iteratively update the current positions of the grey wolf population in the search space. Keep the loop without a break until the end of the IGWO algorithmFinally, the HGAGWO and extreme learning machine (ELM) are combined to construct a classification model (HGAGWO-ELM) for selecting the key features of diabetes and thyroid disease and other disease data. Meanwhile, the IGWO and kernel extreme learning machine(KELM) are also combined to construct a classification model(IGWO-KELM) for the selection of the important features of Parkinson disease and breast cancer and other medical data. In order to verify the proposed methods, a set of metrics are used to evaluate the results of feature selection, including Accuracy, Sensitivity, Specificity, Precision,G-mean and F-measure. The experimental results show that the feature subset selected by HGAGWO-ELM is generally superior to that of GA-ELM and GWO-ELM and other corresponding methods, and the feature subset selected by IGWO-KELM is usually better than that of GA-KELM and GWO-KELM and other homologous methods.