Abstract: At present, the ensemble classification algorithms for incomplete data do not consider the differences among attributes. They weight the sub-classifiers just using the size and the dimension of sub-dataset. In this paper, the information entropy is used to quantify the differences among various sub-datasets, and then the weights for each sub-classifier are computed. So, the weighted voting is fairer, and the prediction accuracy is higher. Experiments on UCI datasets with base classifier of BP show that the proposed algorithm is better than the algorithm using simple weight.

Key words: multi-class AdaBoost, information entropy, incomplete data, ensemble classification