Abstract <p>High-dimensional time series data present challenges in terms of computational costs, storage demands, and noise sensitivity, limiting the efficiency of data mining techniques. Addressing the pervasive challenge of high dimensionality in time series data, a common issue in sensor data analysis, this study introduces a novel solution—the golden-section symbolic approximation (GSSX) representation. Overcoming limitations associated with existing symbolic approaches, particularly the need for predefined user parameters, GSSX leverages the foundations of symbolic aggregate approximation in a two-stage process. Firstly, an adaptive linear space approximation automatically segments time series elements for dimensionality reduction. Additionally, a golden-section-based supervised approach facilitates the discretization process by determining cutpoints based on the actual time series distribution. Experimental validation of GSSX’s applicability includes classification and clustering tasks, demonstrating its ability to retain essential information even with high compression ratios. The study illustrates that GSSX representation competes favorably with state-of-the-art methods. Furthermore, the research investigates GSSX’s robustness through noise and missing data experiments.</p> Graphical abstract <p></p>