Improved Takagi Sugeno Fuzzy Modeling with FLEXFIS-Overlapping Clustering Approach for Efficient Classification Under Conflicts of Interest
S. Anandhavalli, S. K. Srivatsa · International Review on Modelling and Simulations (IREMOS) · 2013
Takagi Sugeno Fuzzy models have been successfully applied to a number of scientific and engineering problems during recent years. A lot of investigations have been done to modify systematic design of Takagi Sugeno fuzzy model. Besides the other, FLEXFIS algorithm is widely used to improve the learning process of Takagi Sugeno model. However, the performance of FLEXFIS algorithm lacks when using the microarray gene expression data i.e. there may be a chance of identifying same data item in two clusters. This will produce the conflict while taking decision about the availability of the data set present in the particular cluster (rule). So to avoid such drawbacks in the existing FLEXFIS algorithm a new technique is proposed in this paper. The proposed method comprises of two stages viz., FLEXFIS with overlapping clustering approach and conflict analysis. The Takagi Sugeno Fuzzy model uses the improved version of FLEXFIS with overlapping clustering approach to cluster the incoming data into the respective clusters. The conflict analysis is performed on the resulting clustered data from the previous stage. The identification and analysis of the data which are present in more than one cluster based on similarities in the behavior is considered. The partitioning of the data into a particular cluster or rule base in a defined manner is carried out in the conflict analysis phase. The implementation result shows the effectiveness of proposed conflict analysis technique in clustering the data in a defined cluster. The performance of the proposed conflict analysis technique is evaluated by conducting different experiments on different microarray gene expression datasets. Moreover, the performance of the proposed technique is compared with the existing FLEXFIS algorithm. The comparison result shows that the proposed technique more accurately clusters the gene data into their corresponding cluster or rule than the existing approach.