Mcdm Based Data Mining Methods Evaluation And Applications

Evaluation of methods or models is a challenging research hot issue in the field of data mining, machine learning, artificial intelligence, as well as management science. No free lunch theorem explicitly states that there exists no single method or model that sould achieve the best performance. However, decision-makers usually pay close attention to the best method or model which can achieve the optimal decision. How to select the best method or model to find the optimal decision based on given problem domains or datasets? In addition, most of researchers are focused on development of new methods or models, without exploring any result in depth, which can result in invisible waste of preliminary knowledge and data resources.For the above stated problems, this paper focuses on the study and discussion of the evaluation of methods or models. The main research contents are as follows:(1) A theoretical framework for the evaluation of methods or models is proposed and developed. Based on multi-criteria decision making, group decision making and data mining, a theoretical framework for the evaluation of methods or models is proposed and developed by combining domain knowledge and expert experience. The theoretical framework consists of three stages and six modules. In particular, the three stages refer to data mining stage, multi-criteria decision making stage and secondary mining stage.(2) An empirical application platform based on the proposed theoretical framework s established. Based on the proposed theoretical framework, an empirical application platform is established for evaluation of classification methods (by the proposed analytic hierarchy model:AHM) and clustering methods (by the proposed consensus decision-making for evaluation of clustering algorithms model:CDMECA), to conduct secondary knowledge discovery in order to enhance the understanding and practicality of mining results.(3) By combining domain knowledge, expert experience with multi-criteria decision making, two improved AHP group decision making methods:IAHP-GDM and EWAHP-GDM are proposed. In this paper, two aggregation methods:aggregation of individual judgments and aggregation of individual priorities are first proposed to unify and integrate in the AHP group decision making. In comparison with the traditional AHP group decision making method, the validity of the two proposed methods are examined and verified by an experimental study. What’s more, the proposed EWAHP-GDM method is extended to determine the criteria weight for further research.(4) A core difficult problem is that different methods may produce different even conflicting results. To address this problem, based on eighty-twenty rule for the secondary mining stage, a consensus reconcilement model is proposed to reconcile disagreements among evaluation performance of clustering methods. In this model, an overall satisfaction of all participants involved in the decision problem can be considered fully and researched quantifiably. In addition, this model can reconcile disagreements among evaluation performance of methods or models.