A visual analytics framework for time-series feature representation and exploration

• Published in: Journal of visualization Vol. 28; no. 5; pp. 1063 - 1082
• Main Authors: Yang, Bin, Zhou, Yixuan, Luo, Xinchi, Qin, Hongxing, Hu, Haibo
• Format: Journal Article
• Language: English
• Published: Heidelberg Springer Nature B.V 01.10.2025
• Subjects: Approximation; Case studies; Datasets; Deep learning; Efficiency; Feature extraction; Machine learning; Representations; Subject specialists; Task analysis; Time series; Trends; Visualization
• ISSN: 1343-8875; 1875-8975
• DOI: 10.1007/s12650-025-01081-6

Time-series data are essential in various fields like environmental meteorology, finance, and health care, as they capture changes and trends over time. However, the increasing amount, dimensionality, and complexity of time-series data pose challenges for traditional analysis methods. Additionally, the visualization and interpretation of such data become more challenging. This paper proposes a visual analysis framework for time-series feature representation and exploration, offering users an integrated approach to their analysis tasks. The framework mainly consists of a deep learning-based time-series feature extraction model and a visual analytics system designed to facilitate time-series analysis. We evaluate the performance of the proposed model using publicly available benchmark datasets, showcasing its superiority with an average classification accuracy of 82.5% compared to other methods. Moreover, the effectiveness of the visual system and the framework is substantiated through case studies with domain experts and a user study involving 19 participants.Graphic Abstract