EUNITE Competition 2002: An N-Tuple Ensemble Classifier Model for Predicting whether a Bank Client is Active or Not

This report presents one selected strategy for dealing with the EUNITE competition task of prediction whether a bank client is active or not. The chosen classification model is based on an ensemble of weak classifiers. Each weak classifier is a simple n-tuple unit, which is probing part of the input attributes and which can deliver as output a vote on one or both classes – active or passive. The report describes the working scenario, regarding both preprocessing of the data and the selection of classification model. Lift factor curves for a left-out validation set are presented and we also describe how the classification model can take part in an adaptive loop, which can modify the model when data change characteristics over time.