Synthetic minority over-sampling technique based on fuzzy c-means clustering for imbalanced data
Hansoo Lee, Seunghyan Jung, Minseok Kim, Sungshin Kim · 2017
Some solutions to solve the class imbalance problem that is one of the representative difficulties in machine learning have been proposed. Among them, SMOTE algorithm is proposed recently to reduce the influence of the problem and shows its remarkable performance to solve real world problems. This paper proposes a novel kind of the SMOTE algorithm by combining the previous SMOTE algorithm and fuzzy logic to deal with uncertainties underlying learning samples. Also, fuzzy c-means clustering is used to assign membership degree to given samples efficiently. Moreover, the extended gap statistics is applied to select the optimal number of cluster. The proposed algorithm is evaluated on using several benchmark datasets and shows its good performance by combining support vector machine classifier.