Optimizing SMOTE by Metaheuristics with Neural Network and Decision Tree

SMOTE (Synthetic minority over-sampling technique) is a commonly used over-sampling technique to subside the imbalanced dataset problem. Traditionally SMOTE has two key important parameters, one is to control the amount of over-sampling, and the other specifies the area of the nearest neighbors. These two parameters are arbitrarily chosen by user. So there are no universally best default values. In this paper, we propose a method that uses metaheuristic optimization algorithms, Bat-inspired algorithm (BAT) and particle swarm optimization algorithm (PSO), to optimize the selection of these two parameters for improving the performance of classifiers for data mining imbalanced data. Users are allowed to define the minimum requirements for two performance indicators, such as Kappa and accuracy. The method iteratively searches for the best pair of SMOTE parameters. Two metaherustics, PSO and BAT are used to find the best parameter values for achieving the required performance via SMOTE. At the end, the highest possible accuracy is obtained while satisfying a minimum degree of Kappa as defined by the user. In comparison to the brute-force method, our method shows advantage in shorter run-time and good classification performance.