Random subspace evidence classifier

摘要: Although there exist a lot of k-nearest neighbor approaches and their variants, few them consider how to make use the information in both whole feature space subspaces. In order address this limitation, we propose new classifier named as random subspace evidence (RSEC). Specifically, RSEC first calculates local hyperplane distance for each class evidences not only space, but also randomly generated Then, basic belief assignment is computed according these distances class. following, all represented by assignments are pooled together Dempster's rule. Finally, assigns label test sample based on combined assignment. The experiments datasets from UCI machine learning repository, artificial data face image database illustrate that proposed approach yields lower classification error average comparing 7 existing variants when performing task. addition, has good performance high dimensional minority imbalanced data.