Scorecard construction with unbalanced class sizes

A long-running issue in scorecard construction in retail banking is how to handle dramatically unbalanced class sizes. This is important because, in many applications, the class sizes are very different. We describe the impact ignoring such imbalance can have and review the various strategies which have been proposed for tackling it, embedding them in a common theoretical framework. We then describe a new ’local’ method of scorecard construction which both theory and our experiments show yields superior performance to standard methods, while retaining their interpretative simplicity. We illustrate using real banking data sets.