A Novel Algorithm for Imbalance Data Classification Based on Genetic Algorithm Improved SMOTE

Abstract

Listen

Translate article

The classification of imbalanced data has been recognized as a crucial problem in machine learning and data mining. In an imbalanced dataset, there are significantly fewer training instances of one class compared to another class. Hence, the minority class instances are much more likely to be misclassified. In the literature, the synthetic minority over-sampling technique (SMOTE) has been developed to deal with the classification of imbalanced datasets. It synthesizes new samples of the minority class to balance the dataset, by re-sampling the instances of the minority class. Nevertheless, the existing algorithms-based SMOTE uses the same sampling rate for all instances of the minority class. This results in sub-optimal performance. To address this issue, we propose a novel genetic algorithm-based SMOTE (GASMOTE) algorithm. The GASMOTE algorithm uses different sampling rates for different minority class instances and finds the combination of optimal sampling rates. The experimental results on ten typical imbalance datasets show that, compared with SMOTE algorithm, GASMOTE can increase 5.9% on F-measure value and 1.6% on G-mean value, and compared with Borderline-SMOTE algorithm, GASMOTE can increase 3.7% on F-measure value and 2.3% on G-mean value. GASMOTE can be used as a new over-sampling technique to deal with imbalance dataset classification problem. We have particularly applied the GASMOTE algorithm to a practical engineering application: prediction of rockburst in the VCR rockburst datasets. The experiment results indicate that the GASMOTE algorithm can accurately predict the rockburst occurrence and hence provides guidance to the design and construction of safe deep mining engineering structures.