Purifying training data to improve performance of multi-label classification algorithms
Sawsan Kanj, Fahed Abdallah, Thierry Denœux · International Conference on Information Fusion · 2012
Multi-label classification assumes that each object in the training set is associated with a set of labels, and the goal is to assign labels to unseen instances. k-nearest neighbors based algorithms answer the multi-label problem by using inherent information given by the neighbors of the observation to classify. Due to several problems, like errors in the input vectors, or in their labels, this information may be wrong and might lead the multi-label algorithm to fail. In this paper, we propose a simple algorithm for editing out some training instances by voting of some metrics in order to purify the existing training sample. This purifying approach is adapted on the recently proposed evidential k-nearest neighbors for multi-label classification. Comparative experimental results on various data sets demonstrate the usefulness and effectiveness of our approach.