Research On Fuzzy Time Series Forecasting Model Based On Fuzzy Decision Tree

In the real world,there are a lot of statistical indicators that are arranged in chronological order,analyzing these data can uncover the inherent laws of the sequence and predict future trends.In classical time series forecasting,historical data is required to have certain integrity and accuracy,but there are many ambiguous facts in real life.According to classical time series cannot deal with the issue of data with uncertainty,Song put forward the concept of fuzzy time series and forecasting model was established.In the process of constructing the fuzzy time series forecasting model,the domain division and the handling of fuzzy relationships are the key to the prediction accuracy.For the domain division,the traditional fuzzy time series forecasting model tend to be divided by experience,there is no basis for domain division,which leads to low prediction accuracy.For the processing of fuzzy relations,a large number of models use single-factor predictions,ignoring other factors related to prediction variables.In addition,many problems in real life are caused by the difficulty of data sampling,data loss,and other reasons that result in missing samples,which makes the model unable to predict relatively accurately.In order to solve these problems,this paper researches and improves the fuzzy time series forecasting algorithm,and puts forward two kinds of forecasting models which based on fuzzy decision tree.For the problem that there is no basis for the division of the traditional fuzzy time series forecasting domain,this paper proposes an adaptive forecasting models with fuzzy-time-series and C-fuzzy decision trees,and optimizes the domain division based on the correlation relationship of sequence data.This model uses C-fuzzy decision tree to mine the association relationship of sequence data,and constructs the objective function through the verification set prediction error and the domain partition complexity,it minimizes the objective function based on the association relationship and adaptively divides the domain for prediction.At the same time,in order to avoid overfitting,based on the decision tree pruning algorithm REP,a C-fuzzy decision tree pruning strategy is proposed for optimization.This paper continues to expand the proposed adaptive fuzzy time series prediction model based on C-fuzzy decision tree,and further proposes a multi-factor model based on random forest.Considering many practical problems,there are not only one influencing factors for predicting variables,at this time,using single-factor predictive models can not meet the needs of the actual situation,and there are a large number of real-world problems with historical data missing.Therefore,using the C-fuzzy decision tree,a variety of factors are introduced to construct the model for prediction,more comprehensive analysis of the characteristics of samples,and through the random forest solve the problem of missing samples.This paper uses simulation experiments to evaluate the model prediction accuracy,applies the model to the prediction of Shanghai Stock Index and TAIEX respectively,and compares the forecast results with other models.According to the experimental results,the model proposed in this paper can overcome the shortcomings of dividing the universe by experience and single-factor prediction,in addition to effectively solve the problem of missing data,high prediction accuracy is obtained in the experiment and the effectiveness of the model is verified.