A potential use of data envelopment analysis for the inverse classification problem

We propose a methodology that uses data envelopment analysis (DEA) for solving the inverse classification problem. An inverse classification problem involves finding out how predictor attributes of a case can be changed so that the case can be classified into a different and more desirable class. For a binary classification problem and non-negative decision-making attributes, we show that under the assumption of conditional monotonicity, and convexity of classes, DEA can be used for inverse classification problem. We illustrate the application of our proposed methodology on hypothetical and real-life bankruptcy prediction data.