Experiments with Adaboost and linear programming
Maria Dolores Valverde Ruiz, Francesc Josep Ferri Rabasa · 2005
In the context of classification problems, boosting refers to methods that create an ensemble of ’weak’ classifiers in order to get a combined classifier that improves the performance of any of the weak classifiers combined and eventually leads to competitive results. It has been reported that the performance of boosting is related to the diversity of the combined classifiers. This paper presents a preliminary approach to obtain a sparse combination of classifiers from a previously learned ensemble that can improve boosting results in such cases.