Feature selection method for binary reptile search algorithm

The computational difficulty of analysing large datasets consisting of data obtained from real-world domains is increasing day by day. Feature selection, which is one of the main methods used to solve this problem, is a very effective method used to reduce unimportant and irrelevant features in the data set. In this study, the Binary Improved Creeper Search Algorithm (BERSA) approach, which combines filter and wrapper feature selection methods for high-dimensional gene and cancer datasets, is proposed in two stages. In the first step, mRMR filtering method is applied to pre-process the data sets and important data are retained. In the second step, 12 different transfer functions were applied for the Creep Search Algorithm, which works in continuous solution space, in order to produce binary values instead of continuous. Thus, 12 different Binary Improved Reptile Search Algorithm (BERSA) variants were produced. Thus, a more detailed study was carried out in the experimental environment. Experiments were conducted using 12 different cancer and gene datasets to measure the performance of the BERSA algorithm. The results of the proposed algorithm are compared with 5 other current metaheuristic algorithms. During the experiments, various performance evaluation metrics such as average accuracy, standard deviation and average number of selected features were taken as basis. The experimental results show that the BERSA algorithm achieves an average accuracy superior to the other compared algorithms. At the same time, it achieved this by selecting fewer features.