Issues with Simple Filters

Simple filters are an easy first step, but these filters easily get it wrong by showing a strong association in the training data, but not the test or new data. These are called “False Positive” variables. Using cross-validation and resampling can reduce carrying False Positive variables forward. The model likely has tuning parameters too, and we end up exploding the models required to run

• I x E x T (Internal resamples x external resamples x tuning parameters)

Fig 11.2