MSc Stats Project Presentation "A Comparison of Random and Rotation Forests"

CANDIDATE:  EBTESAM ALZAHRANI

ABSTRACT:

 The purposes of this project are to investigate and compare two ensemble classification methods, random forest and rotation forest. Both of these methods depend on growing decision trees as base classifier. To achieve these objectives, two datasets are used, Glass Classification dataset and Zoo Animal Classification dataset. To apply these methods on datasets in R, randomForest package and rotationForest package are used. randomForest shows misclassification error, confusion matrix, variable importance and the proximities plot, but rotationForest does not. To predict a new observation, randomForest gives a class directly, but rotationForest gives a probability. The two methods gave the same results in Zoo Animal Classification dataset, but there is a difference in Glass Classification dataset. Also, rotationForest gave an error with small datasets. Overall, we found the randomForest package in R was more useful than rotationForest.