Effects of Varying Parameters in Asymmetric AdaBoost on the Accuracy of a Cascade Audio Classifier

Abstract

Previous work has been done demonstrating that the asymmetric AdaBoost algorithm could be successfully used to design a binary classifier that is a cascade of simple classifiers. We use the algorithm to design an audio classifier and study the effect of changing the length of the cascade and the weighting at each stage on the accuracy of the classifier. Results for a four-class audio classification problem are presented.