Evaluating the quality of test data under the influence of vigilance parameter in flexfis

In this paper, we determine the influence of the vigilance parameter using a modified version of vector quantization used in Flexible Fuzzy Inference System (FLEXFIS) specifically for Takagi Sugeno fuzzy model. FLEXFIS adopts a single pass incremental learning approach for the antecedent parts of the rules' learning process. In order to achieve this learning process, an evolving version of vector quantization is used to either update or evolve new clusters or rules. It helps in the elimination of the outliers (samples with low dense region of the feature space). The use of vigilance parameter steers a tradeoff between plasticity and stability dilemma during the learning process. This is accomplished by selecting the best parameter grid search scenario in association with the cross validation procedure. This ensures some of the desired properties while training the systems during online operational mode such as computational complexity, robustness, preparametrizing of the number of clusters.It also overcomes the problem of cluster projection concept. The adopted algorithm calculates the distance from a new data point to the surface instead of centers as in conventional vector quantization. An evaluation is done on the test data of weather forecasting. A comparative study of the performance analysis for both the conventional and incremental version of vector quantization is also presented in this paper.