Time Series Classification Based on Visualization of Recurrence Plots

Abstract

AbstractOrdered data sets such as time series are found in almost all areas of human activity from cardiograms and to cyberattacks. Classification of time series is one of the most difficult tasks in data mining. In the article, a new method of time series classification based on the construction of recurrence plots is considered. The time series is transformed into a matrix, which characterizes the recurrence of the time series states, and the matrix is presented as a black-and-white image. Further, the convolutional neural network is used to classify the image. The application of the method is demonstrated by examples of simulated time series. A comparative analysis of the classification of noisy time series is carried out. The dependences of the classification accuracy on the noise level of time series are obtained. The results showed that the considered method has a high enough classification accuracy at high noise levels.KeywordsTime seriesNoiseClassificationRecurrence plotConvolutional neural network