Original Source: FICO

Experienced modelers readily understand the value to be derived from developing multiple models based on population segment splits, rather than a single model for the entire population, in any model development project. Analysis of data and model development on subpopulations reveals unique predictive patterns that build greater precision into the segmented model. However, experienced modelers are also aware of the pitfalls in segmentation analysis. It can be extremely time-consuming, and result in over-fitting—segmentation trees with excessive leaf nodes that may or may not add value in predicting the target of interest. A common response to over-fitting, selecting the best and second-best splits to “grow” the segmentation tree, limits the possibility of finding the most promising initial splits, and splits at each node…

Contributors: Open Source
Categories: SIG U Resource, Whitepaper
SRC Type: Risk Management, Sourcing Management, Third Party Management