In this study a CHAID-based approach to detecting classification accuracy heterogeneity across segments of observations is proposed. This helps to solve some important problems, facing a model-builder: 1. How to automatically detect segments in which the model significantly underperforms? 2. How...