Multi-label feature selection based on information entropy fusion in multi-source decision system

Feature selection plays an important role in high-dimensional multi-source data, which can improve classification performance of learning algorithm. Most of existing multi-source information fusion focus on the single decision system without considering multi-source and multi-label problems together. Nevertheless, data from different sources along with multiple labels simultaneously are absolutely frequent in many real-world applications. For this issue, in this paper, a multi-source multi-label decision system is proposed, which has more than one decision label. To remove some redundant or irrelevant features in multi-source multi-label decision system, a feature selection algorithm based on positive region for multi-source multi-label data is explored, which uses the feature dependency carried on the fusion decision table. Finally, examples are introduced to elaborate the detail process of the proposed algorithm, and experimental results show the effective performance of the proposed algorithm on multi-source and multi-label data.