The Improvement Of Fuzzy Time Series Model And Its Application In Unbalanced Data

Since the concept of "fuzzy" has been put forward,many experts and scholars have been put into the relevant research.In nature,people often encounter fuzzy situation is different from either this or that,black and white.The fuzzy time series prediction model based on fuzzy phenomenon is proposed by researchers.Its research value is to solve the problem of uncertainty and fuzziness in traditional time series.After continuous development,fuzzy time series model is applied to all aspects of weather,stock,human flow and so on,and its prediction results are accurate and of high research value.The research focus of fuzzy time series model is mainly on domain partition and sample fuzzification.People combine fuzzy time series and optimization algorithm to get a better prediction result.In this paper,we combine the existing fuzzy time series prediction method and information entropy theory to improve the prediction model of fuzzy time series,and apply it to the prediction of practical problems.In this paper,the main works are as follows:The background and development status of fuzzy time series model are summarized.Then,fuzzy time series prediction model,fuzzy C means clustering algorithm and unbalanced data set are briefly introduced.For the establislhment of fuzzy relation in fuzzy time series model,the concept of entropy is introduced.In the establishment of relational matrix,the original method and information entropy are combined to consider the correlation degree and change trend between classes.We calculate the correlation degree between samples by calculating the value of entropy,adjust the relationship of fuzzy logic dynamically according to the degree of correlation,improve the method of establishing fuzzy relation,and verify the prediction effect of the method through experiment.In the prediction,we will encounter the situation that the data are unbalanced.When the data are predicted,there may be large deviations.In this paper,we combine the improved model with the fuzzy clustering criterion function based on the clustering volume constraint and apply it to the imbalanced data set.Before forecasting,we first use fuzzy C means algorithm to decide the data set.If we do not deal with the unbalanced data,we will process the sample set as the domain,and then use the improved model to predict.Experiments are carried out through real data,new and old models are compared,and a new model is obtained by comparing the minimum standard error values.The error of the prediction results is smaller.