In view of the good classification ability of Moth-Flame Optimization (MFO) in reducing feature redundancy, this paper applied MFO algorithm to feature selection. However, the MFO algorithm is easy to fall into local optimum and has a weak search ability, which severely limits the classification performance and dimensional reduction ability of the algorithm. Therefore, this paper combined MFO algorithm with distributed parallel computing Spark platform distributed, and proposed a feature selection method based on Spark Parallel Binary Moth-Flame Optimization (SPBMFO) algorithm. The experimental results show that compared with the classical particle swarm optimization algorithm(PSO), the genetic algorithm(GA) and the cuckoo search algorithm(CS), when using the binary MFO algorithm for feature selection, the selected features are improved by 12.5%, 15% and 2.5%, respectively. SPBMFO algorithm avoids the search process falling into local optimum and improve the classification performance of the algorithm, which minimizes the number of features while maximizing the classification performance.