| Time series classification is one of the important tasks of time series data mining.Different from other data forms,time series data has the characteristics of real-time updates,high dimensions,and large volumes.The traditional classification method has the problems of high complexity and poor interpretability in processing time series data.Shapelet is a significantly discriminative subsequence of time series,which can effectively provide dimensionality reduction representation and interpretable description of time series.Therefore,the time series classification algorithm based on Shapelet has the advantages of strong interpretability,fast classification speed,and high classification accuracy.This paper focuses on the problems of insufficient extraction accuracy and limited generation mode of the Shapelet-based classification algorithm.The main research contents are as follows:To address the problem of insufficient accuracy in Shapelet extraction,a fast Shapelet discovery algorithm based on subclass clustering and SAX(Symbolic Aggregate Appro Ximation)representation is proposed.This algorithm combines subclass clustering with the classical symbolic representation method SAX to quickly and accurately obtain the optimal Shapelet.The algorithm uses subclass clustering to reduce the dimensionality of time series,obtaining multiple sub-sequence prototypes as Shapelet candidate sets.Then,using SAX to symbolically represent the candidate set,the candidate set is intuitively represented as a string,making it easy to find the optimal Shapelet.Finally,the candidate with the highest information gain is selected as the optimal Shapelet,and time series classification is implemented based on it.Experimental results show that compared with other existing Shapelet discovery algorithms,this algorithm can discover Shapelet more quickly and has higher classification accuracy.In response to the problem of pattern restriction in Shapelet,a time series classification method based on optimized Shapelet is proposed by integrating significant Shapelet learning mechanisms.This method establishes a classification model by combining Shapelet transformation,Laplacian feature mapping,Shapelet similarity measurement,and class label constraints.In the Shapelet candidate set generation phase,K-means clustering is applied to the original time series to overcome the limitation of Shapelet selection only from the original time series,making the learned Shapelet more typical and representative,and accelerating the Shapelet generation process.In the Shapelet selection phase,the model integrates Shapelet similarity measurement to preserve the diversity of Shapelet,thereby enhancing the interpretability of classification.Additionally,class label constraints are incorporated to regulate the differences between training labels and true labels,greatly reducing model errors,and ensuring the efficiency and accuracy of the model.In the Shapelet formation phase,an optimized model strategy is proposed to obtain the optimal Shapelet,which is used as the basis for implementing time series classification.Experimental results show that this method has a high classification accuracy and good classification performance for datasets with significant displacement and distortion features. |
