Topological recursive fitting trees : A framework for interpretable regression extending decision trees
Abstract: Many real-world machine learning applications need interpretation of an algorithm output. The simplicity of some of the most fundamental machine learning algorithms for regression, such as linear regression or decision trees, facilitates interpretation. However, they fall short when facing complex (e.g. high-dimensional, nonlinear) relationships between variables. Several approaches like artificial neural networks and bagging or boosting variants of decision trees have been able to overcome this issue but at the cost of interpretation. We propose a framework called Topological Recursive Fitting (TRF) in which a decision tree is learned based on topological properties of the data. We expect the tree structure of our approach to enable interpretation while achieving comparable performance to previously mentioned blackbox methods. Results show that TRF can achieve comparable performance with those methods, even if confirmation on a more significant number of datasets should be initiated.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)