Question
1
Replies
297
Views
Posted: August 1, 2018
Last activity: August 28, 2018
Closed
Adaptive Models Predictors- Grouping Granularity
While checking the Adaptive Models Predictors performance, I often see a requirement to be able to change the grouping granularity level within specific predictors. For example in a predictor like age maybe grouping granularity as 0.25 is enough, but for the predictor like customer income, I want to increase the granularity to some value so that it increases the number of groups. These predictors coexist within in the same model. Is there a way that I can do this?
Let me clarify how we do the binning. Instead of simple equal size bins (e.g. Income 1000 - 2000, 2000 - 3000 etc) we determine the interval bounds (and the number of intervals) dynamically. The bounds are determined in such a way that the behavior in the intervals is significantly different from eachother. This can result in some interval boundaries being very close to eachother, other intervals being very wide. In you Income example, you would for expect a large interval for all values above a certain limit that captures all people with a very high income (assuming these people have somewhat similar behavior wrt to the outcome that this model is modeling).
For this reason it is seldom needed to change the binning settings, or any other Adaptive Modeling setting for that matter. The general advise is to not change the settings unless you have very good reasons.