Extracting fuzzy measures from sample data: Optimization algorithms and applications
In Multi-Criteria Decision Making (MCDM), decisions are based on several criteria that are usually conflicting and non-homogenously satisfied. We use non-additive (fuzzy) measures combined with the Choquet integral to solve MCDM problems, for they allow to model and aggregate the levels of satisfaction of the several criteria of the problem at hand by considering the relationship between these criteria. In practice, it is difficult to identify such fuzzy measures. An automated process is then necessary and can be used when sample data is available. Several optimization approaches have been used to extract fuzzy measures from sample data, for example, genetic algorithms, gradient descent algorithms, and so on. We propose to automatically model experts' decision process and extract the fuzzy measure corresponding to their reasoning process; to do this, we use samples of the target experts' decision as seed data to automatically extract the fuzzy measure corresponding to the experts' decision process. In particular, we propose several approaches to extract fuzzy measures from sample data, including the Bees algorithm, an adaptive hybrid algorithm that combines the Bees algorithm and an interval constraint solver, and a speculative algorithm. We also apply the proposed approaches to the real data to show the applicability of the algorithms. The real applications include software quality assessment (SQA) and student risk level prediction. Our experimental results show that we are able to improve some of the results obtained through previous approaches, e.g., through machine learning techniques for software quality assessment problem and Cox proportional hazards (PH) regression model for student risk prediction problem.
Wang, Xiaojing, "Extracting fuzzy measures from sample data: Optimization algorithms and applications" (2013). ETD Collection for University of Texas, El Paso. AAI3565946.