A Simultaneous Moth Flame Optimizer Feature Selection Approach Based on Levy Flight and Selection Operators for Medical Diagnosis.

This paper proposes an effective wrapper approach by integrating the Levy flight and evolutionary selection operators into the Moth Flame Optimization (MFO) algorithm. The main purpose is to solve the Feature Selection (FS) problem in medical applications. FS is used as a preprocessing step in a data mining process to improve the performance of the classification system by eliminating irrelevant and redundant features from a dataset. This simplifies the generalization process and reduces the complexity of the generating models. Furthermore, it speeds up the learning process and reduces the cost for additional hardware resources. However, FS is a challenging NP-hard problem because the search space grows exponentially with an increase in the number of features. In this paper, the swarming behavior of the moths is utilized by using the MFO optimizer as a search strategy within a wrapper approach. The Levy flight operator is proposed to enhance the exploratory behavior of the MFO and mitigate the stagnation in local minima. Different selection mechanisms: random selection (RS), tournament selection (TS), and roulette wheel selection (RWS) methods, are investigated to decrease the bias of the MFO algorithm toward exploitation. These selection operators are proposed in the combination of the Levy flight in the form of four different FS methods LBMFO-R1, LBMFO-R2, LBMFO-TS, and LBMFO-RWS. The proposed methods are validated using 23 medical data sets from well-regarded data repositories. The comprehensive results and various comparisons reveal that the Levy flight and selection operators have a great positive impact on the performance of the MFO. They enhance the exploration, convergence trends, and diversity of solutions. [ABSTRACT FROM AUTHOR]