A profit-based performance measure for consumer credit scoring models
Authors: Thomas Verbraken, Wouter Verbeke, Richard Weber, Cristián Bravo and Bart Baesens
Presented at: IADIS European Conference on Data Mining 2012 (ECDM'12) , Lisbon (Portugal), 21- 23 July, 2012
Keywords: Data mining, classification, performance measure, credit scoring
As a result of the steep growth in computational power, the interest for data mining techniques has increased tremendously the past decades. One specific data mining task which is often used is classification, where a categorical target variable is predicted. Typically, the company undertaking the classification exercise will take a certain action based on the outcome of the prediction model. Examples of classification tasks are found in e.g. financial credit scoring, direct marketing response models, fraud detection, and customer churn prediction.
With the increasing number of classification models, model selection becomes more important than ever. Thereby, the focus should be on the end userīs main goal which, in a business setting, usually is profit maximization. Our research aims at implementing a profit-driven classification performance measure specifically for consumer credit scoring models. This performance measure will enable the practitioner to select the credit scoring model which maximizes the profits (or minimizes the losses due to default) of his/her company.