Incremental imputation method for incomplete decision system

Existing imputation algorithms for incomplete decision system are almost non-incremental and rarely consider different contribution to decision label among different features. Therefore, in order to make the most use of information hidden in existing data and reserve the original distribution characteristics, in this paper, we proposes a new incremental imputation algorithm based on attribute significance. Furthermore, in order to overcome the defects of gray correlation degree which is vulnerable to the attribute sequence when searching similar samples, we prefer to map samples to space vectors consisting of all conditional attributes and compute vector similarity form distance and angle. Finally, experiments are tested on several UCI standard datasets and the results show that the proposed imputation algorithm is effective and superior.