Case Study I: Tuning Random Forest (Ranger)

Thomas Bartz-Beielstein, Sowmya Chandrasekaran, Frederik Rehbach, Martin Zaefferer

Abstract: This case study gives a hands-on description of Hyperparameter Tuning (HPT) methods discussed in this book. The Random Forest (RF) method and its implementation ranger was chosen because it is the method of the first choice in many Machine Learning (ML) tasks. RF is easy to implement and robust. It can handle continuous as well as discrete input variables. This and the following two case studies follow the same HPT pipeline: after the data set is provided and pre-processed, the experimental design is set up. Next, the HPT experiments are performed. The R package SPOT is used as a “datascope” to analyze the results from the HPT runs from several perspectives: in addition to Classification and Regression Trees (CART), the analysis combines results from surface, sensitivity and parallel plots with a classical regression analysis. Severity is used to discuss the practical relevance of the results from an error-statistical point-of-view. The well proven R package mlr is used as a uniform interface from the methods of the packages SPOT and SPOTMisc to the ML methods. The corresponding source code is explained in a comprehensible manner.