Summary

Ensemble methods have been found to be very effective for classification, regression, and other related problems. Any statistical and machine learning method must always be followed up with appropriate diagnostics. The assumption that all base models are independent of each other is central to the success of an ensembling method. However, this independence condition is rarely satisfied, especially because the base models are built on the same dataset. We kicked off the chapter with the simplest measure: the geese pair method. With this, we essentially searched for the models that agree with each other at all times. If such models are present in the ensemble, it is safer to remove one of them. With a large dataset and a high number of variables, it is indeed possible that there won't be any base models that speak the same language as another. However, we still need to check whether they are equal. With this in mind, we first proposed measures that compare only two base models...