Combining attribute content and label information for categorical data ensemble clustering

Ensemble clustering has been attracting increasing attention in recent years, because it is able to combine multiple base clusterings (ensemble members) into a more robust clustering. It mainly consists of two parts, generating multiple ensemble members and finding a final partition. The construction of the information matrix plays an important role for finding a final partition. In general categorical data ensemble clustering framework, most existing information matrices are constructed only relying on label information of ensemble members without considering original information of data sets. To solve this problem, a new ensemble clustering framework for categorical data is proposed, in which the information matrix considers label information and original data information together, and is instantiated into the ALM matrix in this paper. The ALM matrix takes account of not only the distribution of attribute content in each ensemble member, but also the relationship among ensemble members based on the distribution. To simplicity, the k-means technique is used to cluster the ALM matrix and form a new ensemble clustering algorithm. The experimental results have shown the benefits of the ALM matrix by comparing the proposed algorithm with other ensemble clustering algorithms.