Some features in a dataset that contain irrelevant or unnecessary data may adversely affect both classification accuracy and the size of data. These negative effects are minimized by using feature selection (FS). Recently, researchers have tried to develop more effective methods by using swarm-based optimization methods in FS, apart from the usual FS methods used in data mining. In this study, a novel wrapper feature selection method based on binary hybrid optimization, called BWPLFS, consisting of a Whale Optimization Algorithm, Particle Swarm Optimization and Lévy Flight is proposed. Ten standard benchmark datasets from the UCI repository for performance evaluation of the proposed algorithm are employed and compared with other literature algorithms. Support vector machines are used both in the objective function of the proposed FS and for classification. The system created for feature selection and classification is run twenty times. As a result of these runs, the average of the fitness values, the average of the classification accuracies, the worst of the fitness values and the best of the fitness values, and the average number of the selected features are found. The BWPLFS is compared with methods in the literature in terms of these criteria. According to the results, it seems that the proposed method selects the most effective features and so it is very promising. In addition, by integrating the proposed algorithm with devices that provide decision support systems, it can be provided to produce more accurate and faster results.