Tree-Based Ensemble Multi-Task Learning Method for Classification and Regression

Multi-task learning is an important area of machine learning that tries to learn multiple tasks simultaneously to improve the accuracy of each individual task. We propose a new tree-based ensemble multi-task learning method for classification and regression (MT-ExtraTrees), based on Extremely Randomized Trees. MT-ExtraTrees is able to share data between tasks minimizing negative transfer while keeping the ability to learn non-linear solutions and to scale well to large datasets.