CLUSplus: A decision tree-based framework for predicting structured outputs

Abstract

We present CLUSplus, a machine learning framework based on decision trees specialized for complex predictive modeling tasks. We provide the scientific community with an open source Java framework that unifies several major research directions in the machine learning field. The framework supports multi-target prediction, i.e., the simultaneous prediction of multiple continuous values, multiple discrete values, and hierarchically organized discrete values. Furthermore, CLUSplus enables state-of-the-art predictive performance via ensemble learning, exploitation of unlabeled data via semi-supervised learning, and data understanding via feature importance and building interpretable models. Out of a wide array of machine learning frameworks available today, very few support complex predictive modeling tasks and, to the best of our knowledge, none support all of the aforementioned functionalities.

Keywords

• Machine learning
• Multi-target regression
• Multi-label classification
• Feature importance
• Semi-supervised learning
• Decision trees