A Study Of A New Method For Fuzzy Rule Weights In Data Classification

Classification is one of the most important research contents in data mining.It is a technology that constructs classification model by analyzing given data sets and classifies samples of known class by using above classification model. The traditional classification methods that deal with well-balanced data can often obtain good classification performance. However, there exists more imbalanced data in the real world. For dealing with classification of imbalanced data, the traditional classification methods often tend to the majority class and lead a lower classification accuracy to the minority class. Thus, it is very important to make researches on the classification methods of the imbalanced data.To different types of data sets, this paper proposes a classification algorithm based on fuzzy rules,which can adjust the classification accuracy on the well-balance data and the imbalanced data effectively. We design a classification algorithm that includes the Chi et al algorithm and the fuzzy reasoning model. It introduces weighting coefficients and pattern distribution function to improve the calculation method of fuzzy rules weights. This algorithm not only keeps the pattern matching degree within class in uniform distribution, but also enhances the contrast of inter-class.Moreover, it weaks the gap of within class and enhances the difference of inter-class. So, classification accuracy is improved by the rules weights adjusted by means of weighting coefficients. In particular, we apply the SMOTE algorithm to preprocess the imbalanced data, which leads to basic balance between the minority class and majority class in quantity. Based on the studies, we classify the imbalanced data by using the classification algorithm. Then, we verify the reliability of the classification algorithm on numerical simulation about the well-balanced and imbalanced data which have the different imbalanced degrees of UCI data sets. In addition, we compare our experimental results with other methods on classification accuracy.Finally, the above results show that the proposed algorithm is superior to other methods.