Abstract

The Boosting algorithm (Schapire, 1990) is extended to fit a wide range of regression problems. This is done via an appropriate theorem on the equivalence of weak and strong learning in a regression framework. The practical capabilities of this model are demonstrated on the laser data from the Santa Fe times series competition, where the results surpass those of standard ensemble average. Keywords: Boosting, Ensemble Averaging, Mixture of Experts, PAC Learning, Regression Estimators, Time series. 1 The Boosting Algorithm Boosting is an ensemble learning algorithm which achieves improved performance by training different learners on different distributions of the data and combining their output (Schapire, 1990). It was suggested in the context of the PAC learning model (Valiant, 1984) and theoretically, it enables achieving arbitrarily low error rate, requiring the basic learners only to have performance which is slightly better than random guessing. The algorithm trains the first lear...

Citations