A Dynamic Classification Method of Time-Series Data Based on Multidimensional Shapelets

Abstract

Time series data classification plays a vital role in the financial analysis of cultural and technological integration enterprises. Still, the current classification of time series data mainly focuses on a single dimension and does not fully consider the dynamics of time series data, resulting in inaccurate classification results. Given the above shortcomings, multidimensional time series are studied, shapelets unit and dimension correlation are defined, and a dynamic classification method of time series data based on multi-dimensional shapelets is proposed. The algorithm screens out the optimal time series for generating shapelets candidate sets from multiple dimensions of the sample, and calculates the correlation coefficient to measure the correlation between the dimensions, and is used to construct the shapelets unit; it is also designed to dynamically update the key parameters to complete double discriminant classification algorithm for classification operation. At the end of the paper, experiments are conducted based on multiple data sets. The experiments show that even in a small sample size, the accuracy of the algorithm designed in this article can reach more than 60%. Furthermore, compared with existing classification algorithms, the classification accuracy of objects with multi-dimensional time-series features is improved by at least 8%.