Abstract: The weight of characteristic attribute index is a significant influence factor during the process of multiple criteria clustering decision aids. Hence,many researches have focused on this important research area. Historical clustering information can effectively provide importance measures for each index with regard to clustering the objects which are to be evaluated. Learning of previous cases can not only contributes to the reveal of the objective law of clustering but also dig out the weight of each attribute index. However, this significant information has been overlooked by many previous researches which can definitely lead to the inaccurate weight calculation. Case learning, in this paper, is defined as the method proposed by the self-reasoning of the results of typical case sets and calculating some of the key parameters, so as to construct the proper decision-making models which can be applied to the evaluation of new objects in the future. To make the most of the existing clustering cases, the objects which are to be clustered as multidimensional attributes are defined by using space vector model. Based on the fact that objects in the same category are more similar than those in different categories, cosine distance is introduced to measure the similarity among different objects. Maximum entropy model is also employed to estimate the expected contribution of different indexes located in diverse levels to the category of the whole object. An illustrative example about weight allocation of attribute indexes in criminal cases is presented in this paper to show how the new approach is applied in the practical clustering decision problem. The feasibility and validity of the newly-proposed method is demonstrated through the comparison analysis with other similar methods. As a decision support, the proposed model can also provide a novel standpoint for weight calculation of objects with multi-level attribute indexes.

Key words: case study, feature space vector, maximum entropy, weight mining