DAG-based Generative Regression

Abstract

Standard regression models address associations between targeted dependent variables and selected independent variables. This paper generalizes this by proposing DAG-based generative regression as a generative process in which the model learns the data generation mechanism from real data. DAG is explicitly involved in the generative process by using structural equation models to capture the data generation mechanisms among the data variables. We learn DAG by reconstructing the model to replicate the real data distribution. We have conducted experiments to measure the performance of our algorithm to show that the results outperform the state-of-the-art by a significantly large margin.