Abstract: In order to ensure the accuracy and diversity of the base classifier at the same time, we proposed an adaptive ensemble algorithm based on clustering and AdaBoost. Firstly, the training samples were divided into multiple clusters by clustering algorithm. Secondly, AdaBoost training was performed on each cluster to get a set of classifiers. Finally, these classifiers were combined according to the weighted voting strategy. The weights of each classifier were adaptive, we calculated the similarity between the test samples and each cluster and got the test samples’ classification confidence given by classifiers. The experimental results show that the algorithm can achieve a higher classification accuracy compared with representative ensemble algorithms such as AdaBoost, Bagging (bootstrap aggregating) and Random Forest.

Key words: clustering, adaptive weight, AdaBoost algorithm, ensemble learning