M4-Greedy: Visualization-Oriented Representative Selection of Time Series Data

To explore time series data patterns, understand data and enable communication, visualization is crucial in various fields. The visual analysis of large-scale time series faces two fundamental challenges: (i) individual series often contain excessive data points that lead to visual clutter when rendered, and (ii) simultaneous display of multiple series results in overlapping patterns that obscure key trends. To address these issues, we propose selecting representative time series that can effectively capture overall trends while maintaining visual clarity. This work investigates the representative selection problem with dual objectives: maximizing both diversity among selected series and their coverage of the complete dataset. We formally prove this bi-objective optimization problem to be NP-hard. Our solution combines a preprocessing technique with a similarity-based greedy algorithm, achieving a 10 \(\times \) speedup with minimal memory overhead across four real-world datasets. Experimental results demonstrate that the proposed method maintains strong performance in both diversity and coverage metrics while significantly improving computational efficiency.