VOTCL and a Case Study of Its Application

In many real-world applications, the problem of class imbalance and cost-sensitive always arise simultaneously. To address this problem, we propose an effective solution named VOTCL: first, we generate several balanced training datasets by combining under-sampling and over-sampling techniques; then, they are trained to get base learners; at last, voting based on optimal threshold is proposed to ensemble those base learners for decision-making. Experiments on the cross-selling dataset provided by PAKDD2007 competition show the effectiveness of our solution with AUC 0.6037.