Classification trees aided mixed regression model |
| |
Authors: | Oguz Akbilgic |
| |
Institution: | 1. Department of Chemical &2. Petroleum Engineering, University of Calgary, Calgary, AB, Canada |
| |
Abstract: | This paper introduces a novel hybrid regression method (MixReg) combining two linear regression methods, ordinary least square (OLS) and least squares ratio (LSR) regression. LSR regression is a method to find the regression coefficients minimizing the sum of squared error rate while OLS minimizes the sum of squared error itself. The goal of this study is to combine two methods in a way that the proposed method superior both OLS and LSR regression methods in terms of R2 statistics and relative error rate. Applications of MixReg, on both simulated and real data, show that MixReg method outperforms both OLS and LSR regression. |
| |
Keywords: | regression least squares ratio regression classification trees prediction |
|
|