Multi-Target Regression Rules With Random Output Selections

In this article, we address the task of multi-target regression (MTR), where goal is to predict multiple continuous variables. We approach MTR by learning global models that simultaneously all target variables, as opposed a separate model for predicting each Specifically, learn rule ensembles generating many candidate rules and assigning them weights are then optimized in order select best performing subset rules. Candidate generated transforming generalized decision trees, called predictive clustering trees (PCTs), into propose extend an existing method named FIRE tree use random output selections (ROS). Such force individual PCTs focus only on randomly selected subsets The obtained from ensemble also various variables (FIRE-ROS). three different methods generate rules: bagging forests PCTs, extremely randomized PCTs. An experimental evaluation range benchmark datasets has been conducted, FIRE-ROS compared interpretable methods, namely rules, original method, well state-of-the-art particular with ROS, linear combinations projections space. results show can improve performance it performs par (non-interpretable) methods.