NNRMLR: A Combined Method of Nearest Neighbor Regression and Multiple Linear Regression

Abstract

To predict the continuous value of target variable using the values of explanation variables, we often use multiple linear regression methods, and many applications have been successfully reported. However, in some data cases, multiple linear regression methods may not work because of strong local dependency of target variable to explanation variables. In such cases, the use of the k nearest-neighbor method (k-NN) in regression can be an alternative. Although a simple k-NN method improves the prediction accuracy, a newly proposed method, a combined method of k-NN regression and the multiple linear regression methods (NNRMLR), is found to show prediction accuracy improvement. The NNRMLR is essentially a nearest-neighbor method assisted with the multiple linear regression for evaluating the distances. As a typical useful example, we have shown that the prediction accuracy of the prices for auctions of used cars is drastically improved.