Two-Class Averaged Perceptron , Two-Class Decision Forest , Two-Class Decision Jungle , Two-Class Logistic Regression
This experiment shows how to build ensemble of heterogeneous classifiers using stacking technique.
The experiment shows how to implement stacking technique for building ensemble of classifiers. This technique was introduced in . We use 4 base classifiers: averaged perceptron, decision forest, decision jungle and logistic regression. Our dataset is CRM upsell dataset from KDD Cup 2009. We would like to find a classifier that optimizes AUC. In the right part of experiment we build 4 base classifiers using the full training set and obtain baseline AUC performance. In the left part of experiment we build a stacked ensemble of classifiers. We split the training set into set 1 and set 2. Set 1 is used to train 4 base classifiers. Set 2 is used to find their best combination. Base classifiers and their ensemble are tuned to optimize AUC. The final results are output of the bottom Execute R module: