Local averaging of heterogeneous regression models

Abstract

Many learning problems involve an exploration of relationships between features in heterogeneous datasets, where different learning models can be more suitable for different regions. We propose herein a technique of localized averaging of regression models. This technique identifies local regions which have similar characteristics and then uses the average value of local experts to describe the relationship between the predictive feature values and the target value. We performed a comparison with other famous combining methods on standard benchmark datasets, and the correlation coefficient of the proposed method was higher.