The Embedded Feature Selection Method Based on Artificial Neural Network Using Gray Wolf Optimizer and Structured Sparsity Norms
Khadijeh Nemati, A. H. Refahi Sheikhani, Sohrab Kordrostami, Kamrad Khoshhal Roudposhti · Research Square · 2023
Abstract Feature selection is important in many machine learning applications. We show that it is not necessary to use all features of a dataset to perform the classification operation and obtain a lower classification error that leads to higher accuracy. We can improve the performance of the proposed method by selecting meaningful features and reducing the dimensions of the features vector. Through the result of our experiment, we show that learning certain feature subsets can lead to better performance on a variety of datasets. The process of finding an adequate number of features is usually time-consuming. Thus, we consider the feature selection problem as an optimization problem and leave the solution to meta-heuristic methods. This paper presents an approach to feature selection by using Gray Wolf Optimizer (GWO) algorithm. The two-layer perceptron as a classifier utilizes the features selected by the GWO algorithm in classification operation. The new sparse norm is used to evaluate the GWO algorithm approach. We call this approach GWO-ANN-SSN. To validate the performance of the GWO-ANN-SSN method, we compared it with existing feature selection methods on some publicly available data sets. In all experimental results, the GWO-ANN-SSN algorithm performs better than other approaches.