Granular Computing Based Fuzzy Modeling Method And Its Applications On Analysis Of Intelligent Data

With the rapid development of modern science and technology, data in real world show explosive growth trends. Data Mining (DM) technology can extract useful knowledge from large amounts of data, and therefore become a hot topic in the field of intelligent data analysis. As a new KDD method, Granular Computing (GrC) mimics human cognition grouping objects which have similar characteristics together to form information granules, then analysis the problem at a higher level based on the appropriate information granules. In this process, the core information can be extracted while the redundant information and the complexity of problem solving are reduced.By taking the above advantages, this paper presents a data driven fuzzy modeling method by using GrC. Fuzzy C-means and K-means method are among the many clustering techniques that have been used to aid the initialization of fuzzy inference system. The main disadvantage of such methods is that the quality of the solutions depends on parameters such as the initial values of the clusters’centers and on estimating the number of clustering. GrC method used in this paper overcomes the disadvantage, it does not require any prior knowledge in the modeling process. By using GrC it is possible to group data together based on similar features. The information granules are always expressed in the form hyperboxes during the information granulation process. In the end of the granulation, the information which can display characteristics of data is extracted from the hyperbox and is used to initialize the fuzzy rule base. Since the parameters of the rule base are all from the data itself, so it has a high transparency and the fuzzy model has strong interpretability. Then the parameters of the fuzzy system are optimized by using the Adaptive Genetic Algorithm (AGA) and Adaptive Neuro-Fuzzy Inference System (ANFIS). And the accuracy of the model is improved. Finally, this method is used for the modeling and analysis of different data sets. The prediction accuracy compared with other methods demonstrates the effectiveness and superiority of the method.