“Rotation Forest: A New Classifier Ensemble Method”

In “Rotation Forest: A New Classifier Ensemble Method“, Rodrıguez and Kuncheva (2006) develop a Random Forest method where the decision boundaries are not of the form   original data feature value <= fixed number. They do this by creating new features by “rotating” the data in the high dimensional feature space before creating a decision tree.  Each decision tree is created with a different rotation of the data, but all of the rotations are generated from Principal Component Analysis. The trees are combined to generate the final classifier.  Here is their abstract

We propose a method for generating classifier ensembles based on feature extraction. To create the training data for a base classifier, the feature set is randomly split into K subsets (K is a parameter of the algorithm) and Principal Component Analysis (PCA) is applied to each subset. All principal components are retained in order to preserve the variability information in the data. Thus, K axis rotations take place to form the new features for a base classifier. The idea of the rotation approach is to encourage simultaneously individual accuracy and diversity within the ensemble. Diversity is promoted through the feature extraction for each base classifier. Decision trees were chosen here because they are sensitive to rotation of the feature axes, hence the name “forest.” Accuracy is sought by keeping all principal components and also using the whole data set to train each base classifier. Using WEKA, we examined the Rotation Forest ensemble on a random selection of 33 benchmark data sets from the UCI repository and compared it with Bagging, AdaBoost, and Random Forest. The results were favorable to Rotation Forest and prompted an investigation into diversity-accuracy landscape of the ensemble models. Diversity-error diagrams revealed that Rotation Forest ensembles construct individual classifiers which are more accurate than these in AdaBoost and Random Forest, and more diverse than these in Bagging, sometimes more accurate as well.