Download or read book Comparing the Accuracy of Flexible Regression Methods written by Emre Ipek and published by . This book was released on 2014 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: This thesis compares the accuracy of different statistical and machine learning methods by means of a simulation study. The procedures that are tested can be classified in three different categories: regularized linear methods (subset selection, the lasso and ridge regression), flexible methods (smoothing splines, generalized additive models, classification and regression trees, multivariate adaptive regression splines and neural networks) and variance and bias reduction techniques (bagging and boosting). The simulation study includes five different data generating models, which vary in their characteristics concerning linearity and additivity, with every of the models being evaluated for a large and a small sample size. The results of this study do not provide an "off-the-shelf" method, which performs best in all situations. Nonetheless, it suggests that bagging is the procedure that should generally be favored, yielding the best prediction results in nine out of ten cases. However, the strength of bagging comes at a price, as it is the computationally most intensive method and leads to a loss of interpretability.