|
Abstract:
|
Unlike ten or so years ago, most managers would agree that database marketing statistical models provide significant advantages. While this, in itself, is certainly an encouraging sign, model evaluation is not as simple a matter as many would believe. Many ingredients of the evaluation process are overlooked. Some believe that with the presence of a gains or performance report, the evaluation criteria are obvious. The standard frequently summoned is "does the decile report seem reasonable?" While many managers and analysts would walk away happy with viewing a monotonically decreasing gains report, there are additional dimensions that can impact model performance, but are frequently ignored by both marketers and data miners. These include percent captured variations, choppiness, stability, predictability, and explainability.
|
