A clustering based method to complete frame of discernment

When the existing information does not contain all categories, Generalized Evidence Theory (GET) can deal with fusion. However, question of how to determine number categories through GET is still intriguing. To address this question, a modified k-means clustering, named centers initialized clustering proposed, filling gap identification and complement frame discernment. Based on method, determined. The selected by center density keep cluster results constant, enhancing stability results. Besides, constructing basic Probability Assignment (GBPA) modules in conservative way improves reliability mass empty set combined GBPAs indicator categories. Experiments real artificial data sets are conducted show effectiveness.