A hybridization of multiple imputation and one-class bagging ensemble approach for missing value and class imbalance problem

Class imbalance in a dataset leads to erroneous outcomes that engrave the learning techniques and high misclassification cost in the minority class. Along with class imbalance, missing values present in a disproportionate amount in a dataset create a great hindrance to the effective performance of a method. The ensemble method, where multiple methods are ensembled, tackles such issues and shows good results compared to the performance of individual methods. In this paper, a hybridization of multiple imputation and one-class bagging ensemble approach is proposed that handles datasets having both class imbalance and missing values. An in-depth analysis of this approach is studied and the effectiveness of the class imbalance is also presented. To tackle the misclassification of minority samples and missing values, factor-based multiple imputation oversampling technique is used and one-class classifier is ensembled to increase the performance of the class imbalance datasets. Experiments are performed using a one-class support vector machine classifier and the results are evaluated using metrics: Recall (Detection rate), Specificity, f-measure, g-mean, AUC, and Precision. The proposed approach yields a 6.3% improvement in Recall, whereas Specificity, f-measure, g-mean, AUC, and Precision show that the proposed approach improves by 4.92%, 11.3%, 9.4%, 8.3%, and 8.03%, respectively.