GAB-EPA: A GA Based Ensemble Pruning Approach to Tackle Multiclass Imbalanced Problems

摘要

Processing imbalanced data sets has become a challenging issue in machine learning and data mining communities. Although many researches in the literature have focused on two class problems, multiclass problems have attracted a lot of attention recently. Many existing solutions for multiclass tasks are focused on class decomposition methods, i.e. divide the problem into some two-class sub-problems which are easier to handle. This paper presents a Genetic Algorithm-Based Ensemble Pruning Algorithm, called GAB-EPA, for multiclass imbalanced problems without applying any class decomposition techniques. In effect, GAB-EPA seeks to find the best subset of classifiers that not only are accurate in their predictions, but also can generate an admissible diversity when gather together as an ensemble model. To show the effectiveness of our approach, we compared our results with other popular ensemble algorithms in terms of three evaluation metrics: Minority Class Recall, G-mean, and MAUC.