Fusion Regression

Abstract

In recent years, various regression methods have been studied in the literature. Although these methods have shown success in different applications, there is no consensus on which one is the best. Different regressors can produce significantly different prediction results when applied to datasets with varying properties. In this paper, we propose Fusion Regression (FuR), a novel approach that combines the predictions of multiple regressors to leverage their complementary views. FuR concatenates the predictions of regressors to create a new feature space and employs a re-ranking scheme for improved accuracy. Our experiments, conducted on 10 datasets with varying properties (such as size and dimension), show that FuR leads to performance gains of up to 20% compared to the best baseline regressor and up to 16% compared to the recently proposed Regression by Re-ranking method.