Developing a classification Method for Imbalanced Dataset Using Multi-Objective Evolutionary Algorithms

Data classification is one of the main issues in management science which took into account from different approaches. Artificial intelligence methods are among the most important classification methods, most of them consider total accuracy function in performance evaluation. Since in imbalanced data sets this function considers the cost of prediction errors as a fix amount, in this research a sensitivity function in used in addition to the accuracy function in order to increase the accuracy in all of the predefined classes. In addition, due to complexity in process of seeking information from decision maker, NSGA II algorithm is used to extract the parameters (Weight vector and cut levels between classes). In each iteration, based on the estimated weight vector and data sets, the algorithm calculate the score of each alternative using Sum Product function and then allocates the alternative to one of the classes, comparing to the estimated cut levels,. Then, using the fitness functions, the estimation class and the actual class will compare by two algorithms and this process will continue since optimizing the parameters. Comparison of the NSGA II and NRGA algorithms show the high efficiency of the proposed algorithm.