BFRA: A New Binary Hyper-Heuristics Feature Ranks Algorithm for Feature Selection in High-Dimensional Classification Data

Feature selection is one of the main issues in machine learning algorithms. In this paper, a new binary hyper-heuristics feature ranks algorithm is designed to solve the feature selection problem in high-dimensional classification data called the BFRA algorithm. The initial strong population generation is done by ranking the features based on the initial Laplacian Score (ILR) method. A new operator called AHWF removes the zero-importance or redundant features from the population-based solutions. Another new operator, AHBF, selects the key features in population-based solutions. These two operators are designed to increase the exploitation of the BFRA algorithm. To ensure exploration, we introduced a new operator called BOM, a binary counter-mutation that increases the exploration and escape from the BFRA algorithm's local trap. Finally, the BFRA algorithm was evaluated on 26 high-dimensional data with different statistical criteria. The BFRA algorithm has been tested with various meta-heuristic algorithms. The experiments' different dimensions show that the BFRA algorithm works like a robust meta-heuristic algorithm in low dimensions. Nevertheless, by increasing the dataset dimensions, the BFRA performs better than other algorithms in terms of the best fitness function value, accuracy of the classifiers, and the number of selected features compared to different algorithms. However, a case study of sentiment analysis of movie viewers using BFRA proves that BFRA algorithms demonstrate affordable performance.