Selecting Accurate and Comprehensible Regression Algorithms through Meta Learning

Data mining tools often include a workbench of algorithms to model a given dataset but lack sufficient guidance to select the most accurate algorithm according to the nature of the dataset. The most accurate algorithm is not known in advance and no single model format is superior for all datasets. An a priori comparison experiment to determine which algorithm leads to the most optimal model fit could offer relief but is rather time consuming. A meta model which is able to predict and explain which model would fit best could address this problem. The relevance of such a meta model increases with the number and runtime of algorithms under consideration and the size of the dataset. In this paper a novel meta model is proposed to automatically provide support to the user about whether to use a linear, spline, tree, linear tree or linear spline model, given a particular dataset. This study distinguishes itself from previous meta learning studies by focusing on comprehensible regression models, regression specific dataset characteristics, artificially generated datasets and a score based user recommendation.