ISMOTE Oversampling Algorithm For Imbalanced Data Classification

In recent years, the problem of imbalanced classification has been widely concerned. The SMOTE algorithm can expand the minority class samples and improve the classification ability of the minority class in the imbalanced data set, but there is a certain blindness in the selection of minority class sample points and the selection of random numbers. In this paper, the I-SMOTE algorithm (Improved Synthetic Minority Oversampling Technique) is proposed. The algorithm gives the definition of noise and processes the noise. According to the distribution law of minority samples, minority samples can be synthesized more reasonably. The interpolation correlation coefficient is carefully divided when synthesizing samples. Four UCI unbalanced data sets were selected and three classical single classifier algorithms were used for experiments. The SMOTE algorithm and the ISMOTE algorithm were compared. The results showed that the classification results of the ISMOTE algorithm were improved in different degrees in F value, G-mean value and AUC value, which proved the effectiveness of the algorithm.