Time Series Classification Based on Intuitionistic Fuzzy Clustering Similarity Measure

At present, the method of using conventional distance to calculate the similarity between sequences needs to be further improved. It seems challenging to accurately characterize the shape and trend changes of time series solely based on analyzing and computing the time series data itself. In this paper, fuzzy theory is introduced to analyze time series and reconstruct high-dimensional time series. The reconstructed segmented time series encapsulates the trend changes throughout the entire time span. By defuzzifying the reconstructed segmented time series, it is mapped to the corresponding morphological representation, yielding a representation vector that reflects the morphological shape and trend changes of the time series. Furthermore, the similarity of the new representations is measured in combination with dynamic time warping, and the new representations are classified based on the calculated distances. This paper proposes a classification algorithm based on the intuitionistic fuzzy clustering similarity measure, which measures the variation trend of time series from the perspectives of global approximation and local difference. Experimental validation on real-world datasets demonstrates the effectiveness and high classification accuracy of our proposed method.