A Multi-Threshold Granulation Model for Incomplete Decision Tables

How to establish basic granules of knowledge is a fundamental issue for data mining from incomplete decision tables. In the existing methods, basic granules under similarity relation contain too many objects and disturb the later knowledge mining, while granules under limited similarity relation, although simplifying the granules through introducing a limited threshold on two objects satisfying similarity relation, still have problems such as high computation and low prediction precision. In this paper, a multi-threshold model is presented to establish basic knowledge units of incomplete decision table based on the idea of granular computing, comparison experiments on the new model with two existing models show that the new model is superior to the other models on prediction precision, time cost and attribute reduction.