Time Series Classification Based On Temporal Fuzzy Cognitive Map

Time series is the observation value obtained by observing the variables in a system in chronological order and recording in unit time.According to the number of variables observed in the system,the time series can be divided into multivariable time series(MTS)and univariate time series.MTS classification is an active research area in MTS research.Its research task is to establish a classifier and map the input MTS to its corresponding classification label.In the classification research of MTS,considering the characteristics of MTS in spatial dimension and time dimension can improve the classification accuracy.In addition,in some studies of MTS classification,it is necessary to obtain interpretable classification results.For example,in the prediction of financial distress,an enterprise is classified as a financial distress enterprise,and business operators or investors want to obtain the corresponding reasons.Existing time series classifiers mainly include general machine learning classifiers and deep neural network classifiers.Studies have shown that the classification of MTS using general machine learning classifiers does not consider the relationship between different variables in the time dimension,and the accuracy of the classification results is often not high;the classification results of deep neural network classifiers have high accuracy,but because of Its complex structure is regarded as a black box and cannot explain the reason for the classification result.The temporal fuzzy cognitive map(tFCM)is a time series prediction model of fuzzy cognitive map(FCM)extended in the time domain.It uses the relationship between the cross-dimensional dimensions of variables to perform causal relationship inference in the time domain.Each variable produces accurate and interpretable predictions,but cannot be classified.Therefore,this paper introduces the module evaluation node and the system evaluation node into the model,and maps the output of the model to the category label of MTS to construct the MFC classification model based on tFCM.In this thesis,the MTS classification model based on tFCM is constructed in four steps.First of all,according to the actual application scenarios of the system,the variables in the MTS are divided into modules,and different modules represent different performances of the system.The module is evaluated;then,the evaluation results of all modules at the same time are integrated to evaluate the system at that time;finally,the results of the system evaluation at the latest multiple times are integrated to produce a classification result.In order to verify the accuracy of the model,this paper conducted experiments on the m Healthy time series data set,EMA data set and listed company financial time series data set,and will be combined with the vector machine(SVM),decision tree(DT),Encoder and FCN models comparing.Experimental results show that in the m Healthy dataset,when the sequence length is 7 < T < 25,theclassification accuracy of the tFCM model is on average higher than that of SVM and DT by5.86% and 5.08%,respectively,and the accuracy is the same as the Encoder model quiteIn order to further verify the interpretability of the MTS classification model based on tFCM,this paper constructs the MTS classification Web service system based on tFCM,and takes the case of corporate financial distress classification as an example to expand the case analysis.The article first introduces the structural design of the system and the realization of the system prototype,then introduces the related concepts of the enterprise financial time series,and uses the system to establish the tFCM-based classification model of enterprise financial distress,and then combines the knowledge of enterprise financial management to the right Value analysis,put forward suggestions to modify the model weights,and finally through a comparative analysis of the classification results of the financial distress of multiple companies,explain why these companies are classified as financial distress or non-financial distress companies,this shows the interpretability of the model.