Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
Matrix Profile based Shapelets Discovery for Time Series Classification
1Sichuan Provincial Key Laboratory of Philosophy and Social Science for Language Intelligence in Special Education, Leshan Normal University, Leshan Sichuan, China.
2School of Computer Science, Southwest Petroleum University, Chengdu Sichuan, China
3Key Laboratory of Internet Natural Language Processing of Sichuan Provincial Education Department, Leshan Normal University, Leshan Sichuan, China
ABSTRACT
Shapelet is a continuous subsequence that best represents a certain category of time series. Due to the interpretability and high accuracy of shapelet, numerous researchers have been attracted to study shapelet-based time series classification (S-TSC) in the past decade. However, most S-TSC methods search for shapelets on a complete time series, generating a large number of candidate shapelets and resulting in significant time consumption in shapelets evaluation. In this work, we propose a shapelet discovery method based on Matrix Profile for dealing with multi-class time series classification. Firstly, we select 3 subsequences in each category of time series using the K-means clustering. Secondly, we use Matrix Profile to find a number of critical regions from which the shapelets are extracted. Finally, the shapelets are used for the transformation of the original time series which produces a new data sets. The proposed method can significantly reduce the number of candidate shapelets, thus it improves the efficiency of the S-TSC algorithm. Experiments show that the proposed method can improve the efficiency of shapelets discovery, while maintaining high accuracy in multi-class time series classification.
Keywords: Time series, Multi-class time series classification, Shapelet discovery, Matrix profile.
Download PDF
1Sichuan Provincial Key Laboratory of Philosophy and Social Science for Language Intelligence in Special Education, Leshan Normal University, Leshan Sichuan, China.
2School of Computer Science, Southwest Petroleum University, Chengdu Sichuan, China 3Key Laboratory of Internet Natural Language Processing of Sichuan Provincial Education Department, Leshan Normal University, Leshan Sichuan, China
ABSTRACT
Shapelet is a continuous subsequence that best represents a certain category of time series. Due to the interpretability and high accuracy of shapelet, numerous researchers have been attracted to study shapelet-based time series classification (S-TSC) in the past decade. However, most S-TSC methods search for shapelets on a complete time series, generating a large number of candidate shapelets and resulting in significant time consumption in shapelets evaluation. In this work, we propose a shapelet discovery method based on Matrix Profile for dealing with multi-class time series classification. Firstly, we select 3 subsequences in each category of time series using the K-means clustering. Secondly, we use Matrix Profile to find a number of critical regions from which the shapelets are extracted. Finally, the shapelets are used for the transformation of the original time series which produces a new data sets. The proposed method can significantly reduce the number of candidate shapelets, thus it improves the efficiency of the S-TSC algorithm. Experiments show that the proposed method can improve the efficiency of shapelets discovery, while maintaining high accuracy in multi-class time series classification.
Keywords: Time series, Multi-class time series classification, Shapelet discovery, Matrix profile.
Download PDF