Extrapolation errors in linear model trees

Authors:
Wei-Yin Loh;Chien-Wei Chen;Wei Zheng
Affiliations:
University of Wisconsin, Madison, WI;University of Wisconsin, Madison, WI;University of Wisconsin, Madison, WI
Venue:
ACM Transactions on Knowledge Discovery from Data (TKDD)
Year:
2007

Citing 7
Cited 1

Multivariate data analysis with readings (2nd ed.)

Multivariate data analysis with readings (2nd ed.)
Practical Data Analysis: Case Studies in Business Statistics

Practical Data Analysis: Case Studies in Business Statistics
Random Forests

Machine Learning
Statistical Analysis: A Computer Oriented Approach

Statistical Analysis: A Computer Oriented Approach
Regression Modeling Strategies

Regression Modeling Strategies
Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)

Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)
Rule-based machine learning methods for functional prediction

Journal of Artificial Intelligence Research

Data mining and model trees study on GDP and its influence factors

AIASABEBI'11 Proceedings of the 11th WSEAS international conference on Applied informatics and communications, and Proceedings of the 4th WSEAS International conference on Biomedical electronics and biomedical informatics, and Proceedings of the international conference on Computational engineering in systems applications

Quantified Score

Hi-index	0.00

Visualization

Abstract

Prediction errors from a linear model tend to be larger when extrapolation is involved, particularly when the model is wrong. This article considers the problem of extrapolation and interpolation errors when a linear model tree is used for prediction. It proposes several ways to curtail the size of the errors, and uses a large collection of real datasets to demonstrate that the solutions are effective in reducing the average mean squared prediction error. The article also provides a proof that, if a linear model is correct, the proposed solutions have no undesirable effects as the training sample size tends to infinity.