![]() ![]() Sometimes, however, we may be more interested in cases with the largest prediction errors, which can be identified with the help of residuals. In Chapter 15, we discussed measures that can be used to evaluate the overall performance of a predictive model. If we find any systematic deviations from the expected behavior, they may signal an issue with a model (for instance, an omitted explanatory variable or a wrong functional form of a variable included in the model). The single-instance explainers can then be used in the problematic cases to understand, for instance, which factors contribute most to the errors in prediction.įor most models, residuals should express a random behavior with certain properties (like, e.g., being concentrated around 0). Residuals can be used to identify potentially problematic instances. In Part II of the book, we discussed tools for single-instance exploration. The methods may be used for several purposes: In particular, we focus on graphical methods that use residuals. In this chapter, we present methods that are useful for a detailed examination of both overall and instance-specific model performance.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |