Research On The Multi-dimensional Time Series Classification Based On Shapelets

In most scientific fields,data measurements are performed over time.These observations allow people to collect a series of ordered data,called time series.Time series classification is an important research content of time series data mining.With the reduction of hardware costs such as sensors,people often collect different parameters of the same thing for analysis,which brings multi-dimensional time series.In recent years,time series shapelets have attracted a great deal of attention.It has achieved the purpose of accurately classifying time series by identifying the local features of sequential data.Most researches on shapelets are focused on onedimensional time series.However,there are few studies on the classification of multidimensional time series using shapelets.In the study of one-dimensional time series classification methods using shapelets,for the time series with trend change,the typical time series representation is used for shapelets discovery,which tends to cause the loss of trend information of time series.Aiming at the loss of time-series trend information,this paper presents a new trend-based diversified Top-k shapelets classification method(TDTS).By using the method of trend feature symbolization,TDTS algorithm can guarantee the classification effect of the sequence and the efficiency of shapelets discovery while preserving the trend information of time series data.Experimental results of time series classification show that compared with the traditional classification algorithm,the accuracy of the proposed method was improved on 11 experimental data sets;compared with Fast Shapelet algorithm,the efficiency was improved,the running time of the proposed method was shortened,specially for the data with obvious trend information.Aiming at the problem of few research on the concept of shapelets in multidimensional time series classification,this paper presents a multi-dimensional time series classification method based on TDTS.The algorithm uses the idea of bagging in ensemble learning to enhance the generalization performance of ensemble learning by enhancing the diversity of individual learners.The paper conducts experiments with four perspectives: comparison with traditional classification algorithms,comparison with each single dimension classification result,comparison with the time series classification based on diversified top-k shapelets,and comparison with Shapelet Ensemble algorithm.Experimental results show that the proposed algorithm can effectively improve the classification effect of multi-dimensional time series while preserving the trend of data.Finally,on the basis of the theoretical research of this paper,a modular development method is used to design and implement a prototype system for onedimensional and multi-dimensional time series classification,which makes the system clearly display time series classification results and other related information.Furthermore,the function can also easily verify the validity of the proposed method.