Why CART Works for Variability-Aware Performance Prediction? An Empirical Study on Performance Distributions

TitleWhy CART Works for Variability-Aware Performance Prediction? An Empirical Study on Performance Distributions
Publication TypeReport
Year of Publication2013
AuthorsGuo, J., K. Czarnecki, S. Apel, N. Siegmund, and A. Wąsowski
Date Published04/2013
InstitutionGenerative Software Development Laboratory, University of Waterloo
CityWaterloo
TypeTechnical Report
Report NumberGSDLAB-TR-2013-04-02
Abstract

This report presents follow-up work for our previous technical report "Variability-Aware Performance Modeling: A Statistical Learning Approach" (GSDLAB-TR-2012-08-18). We try to give evidence why our approach, based on a statistical-learning technique called Classification And Regression Trees (CART), works for variability-aware performance prediction. To this end, we conduct a comparative analysis of performance distributions on the evaluated case studies and empirically explore why our approach works with small random samples.

AttachmentSize
GSDTR20130402gjm.pdf450.96 KB