AdaBoost.RT: a boosting algorithm for regression problems

• Published in: 2004 IEEE International Joint Conference on Neural Networks (IEEE Cat. No.04CH37541) Vol. 2; pp. 1163 - 1168 vol.2
• Main Authors: Solomatine, D.P., Shrestha, D.L.
• Format: Conference Proceeding
• Language: English
• Published: Piscataway NJ IEEE 2004
• Subjects: Accuracy; Applied sciences; Artificial intelligence; Artificial neural networks; Bagging; Boosting; Classification algorithms; Computer science; control theory; systems; Error correction; Estimation error; Exact sciences and technology; Filtering; Machine learning; Training data; Estimation error; Filtering; Error estimation; Supervised learning; Classification; Relative error; Regression analysis; Neural network; Artificial intelligence; Modeling; Aggregate model
• ISBN: 0780383591; 9780780383593
• ISSN: 1098-7576
• DOI: 10.1109/IJCNN.2004.1380102

A boosting algorithm, AdaBoost.RT, is proposed for regression problems. The idea is to filter out examples with a relative estimation error that is higher than the pre-set threshold value, and then follow the AdaBoost procedure. Thus it requires to select the sub-optimal value of relative error threshold to demarcate predictions from the predictor as correct or incorrect. Some experimental results using the M5 model tree as a weak learning machine for benchmark data sets and for hydrological modeling are reported, and compared to other boosting methods, bagging and artificial neural networks, and to a single M5 model tree. AdaBoost.Rt is proved to perform better on most of the considered data sets.