Modeling Transitions of Inter-segment Patterns for Time Series Representation

Against the backdrop of technological advancements, we are now equipped to collect and analyze time series data in unparalleled ways, offering significant value across various fields. However, traditional time series data analysis often leans heavily on expert insight. This study introduces a novel approach to time series data analysis based on the shapelet evolution graph, designed to intuitively capture core patterns and characteristics within the data without the need for expert intervention. Comparative analysis reveals that our approach excels in scenarios with explicit pattern transitions. Our research not only offers a fresh perspective and methodology for time series data analysis, through comparison with other baseline methods, but also provides foundational knowledge to predict whether a dataset exhibits pattern transition phenomena.