Cost-Sensitive Modeling to Predict Bank Telemarketing Campaign

Building classification models for prediction is a critical task in data mining applications. The task becomes very challenging when datasets exhibit class imbalance and the application at hand is very sensitive towards inaccurate prediction. Normally, classification models tend to be overwhelmed by the majority class instances and ignore minority class instances in the imbalanced dataset, resulting in misclassifications and inaccurate predictions. In many applications, the cost of misclassification is different for different classes. Consequently, the model becomes inefficient for prediction. Cost-sensitive learning considers different costs of misclassification and improves model performance. The goal is to minimize the total cost of misclassification while making effective prediction. This paper presents a comparative study of cost-sensitive classification and prediction approaches considering a telemarketing application