Stacked regression with a generalization of the Moore-Penrose Pseudoinverse

Abstract

In practice, it often happens that there are a number of classification methods. We are not able to clearly determine which method is optimal. We propose a combined method that allows us to consolidate information from multiple sources in a better classifier. Stacked regression (SR) is a method for forming linear combinations of different classifiers to give improved classification accuracy. The Moore-Penrose (MP) pseudoinverse is a general way to find the solution to a system of linear equations. This paper presents the use of a generalization of the MP pseudoinverse of a matrix in SR. However, for data sets with a greater number of features our exact method is computationally too slow to achieve good results: we propose a genetic approach to solve the problem. Experimental results on various real data sets demonstrate that the improvements are efficient and that this approach outperforms the classical SR method, providing a significant reduction in the mean classification error rate.