A novel random forest approach using specific under sampling strategy

Abstract

In Data Mining the knowledge is discovered from the existing real world data sets. In real time scenario, the category of datasets varies dynamically. One of the emerging categories of dataset is class imbalance data. In Class Imbalance data, the percentages of instances in one class are far greater than the other class. The traditional data mining algorithms are well applicable for knowledge discovery from balance datasets. Efficient knowledge discovery is hampered in the case of class imbalance datasets. In this paper, we propose a novel approach dubbed as Under Sampling using Random Forest (USRF) for efficient knowledge discovery from imbalance datasets. The proposed USRF approach is verified on the 11 benchmark datasets from UCI repository. The experimental observations show that an improved accuracy and AUC is achieved with the proposed USRF approach with a good reduction in RMS error.