Identifying Conditionally Independent Target Subsets for Multi-Target Regression
Orhan Yazar, Haytham Elghazel, Mohand-Saïd Hacid, Nathalie Castin · 2020
Multi-target regression MTR is a challenging task that consists of creating predictive models to predict multiple continuous targets simultaneously. The benefit of exploiting conditional target dependencies in MTR can in principle improve the generalization performance but is known to be closely dependent on the properties of the data and the type of loss to be minimized. In MTR data where many inter-dependencies between the targets may be present, explicitly modeling all inter-target and input-output relationships is intuitively far more reasonable. In this paper, we address this problem by designing a MTR approach based on Bayesian networks and demonstrate that using the Markov blanket graph of the multiple target variables to explicitly identify different target powersets and their optimal set of predictors can serve as a powerful learning framework for MTR. Experimental results on various benchmark MTR data sets approved that the proposed method enjoys significant advantages compared to other state-of-the-art MTR methods.