A New Method Of Time-Series Event Prediction Based On Sequence Labeling

Time series event prediction(TSEP)is one of the main objectives of time series mining.It has always been an active research field and plays an important role in appropriate decision-making in different application fields.In the existing TSEP method research,most of the work focuses on improving the classification algorithm of subsequence sets(sets composed of multiple adjacent subsequences).However,these prediction methods ignore the time dependence between subsequence sets,and do not capture the mutual conversion relationship between events,so the prediction effect of small sample data sets is very poor.Sequence marking is one of the common problems in natural language processing and image segmentation.Inspired by the sequence marking problem,this paper proposes a new time series event prediction framework,which transforms the event prediction problem into a marking problem to better capture the temporal relationship between the subsequence sets.The main work of this paper is as follows:First,the theory of time series event prediction is reviewed.This paper introduces the definition and theoretical basis of event prediction,studies and analyzes two kinds of traditional time series event prediction methods,and summarizes the existing prediction framework? For each part of the framework,several common algorithms are introduced and their advantages and disadvantages are analyzed.Secondly,the event prediction method based on sequence annotation(CX-LC)is proposed.Specifically,the framework uses sequence clustering algorithm to identify representative patterns in time series for the first time,and then represents the subsequence set as a weighted combination of patterns,and uses the limit gradient enhancement algorithm(XGBoost)for feature selection.After that,the selected pattern features are used as input to the long-term short-term memory model(LSTM)to obtain preliminary prediction values.In addition,the full link conditional random field(CRF)is used to smooth and refine the preliminary prediction value to obtain the final prediction result.Third,algorithm implementation.The experimental results of event prediction for five real data show that the CX-LC method has a certain improvement in prediction accuracy compared with the other nine models.