TSExplorer: An interactive data annotation and exploration tool for time-series data

A cross-platform, general-purpose research tool for visualizing, annotating, and refining high-dimensional time-series datasets

Einari Vaaras; Manu Airaksinen; Okko Räsänen· 2026· DOI 10.48550/arXiv.2608.30514

The core problem

Time-series data is ubiquitous across domains such as healthcare, finance, sensor networks, and human behavior analysis, yet working with such data remains challenging due to its high dimensionality, temporal dependencies, and frequent lack of reliable labels. Existing tools often focus on either visualization or annotation, but rarely integrate both into a single, general-purpose workflow. TSExplorer addresses this gap by providing a cross-platform environment for interactive annotation and exploration of time-series data. The tool is designed to support a wide range of research workflows, including exploratory data analysis, annotation of unlabeled or partially-labeled datasets, comparison of feature representations, and post-hoc inspection and refinement of existing labels with interactive visual feedback. By combining multiple complementary 2D visualizations derived from high-dimensional feature representations, TSExplorer enables users to reason about complex temporal data without requiring deep expertise in dimensionality reduction or visualization engineering.

The abstract reports that TSExplorer enables users to inspect high-dimensional datasets through multiple complementary 2D visualizations derived from high-dimensional feature representations. It is presented as a cross-platform tool that supports a wide range of workflows. Specific quantitative results, user studies, or benchmark comparisons are not provided in the source material. The primary result is the tool itself: a general-purpose research instrument for interactive annotation and exploration of time-series data. Its capabilities include exploratory data analysis, annotation of unlabeled or partially-labeled datasets, comparison of feature representations, and post-hoc inspection and refinement of existing labels with interactive visual feedback.

Time-series data is ubiquitous across domains such as healthcare, finance, sensor networks, and human behavior analysis, yet working with such data remains challenging due to its high dimensionality, temporal dependencies, and frequent lack of reliable labels. Existing tools often focus on either visualization or annotation, but rarely integrate both into a single, general-purpose workflow. TSExplorer addresses this gap by providing a cross-platform environment for interactive annotation and exploration of time-series data. The tool is designed to support a wide range of research workflows, including exploratory data analysis, annotation of unlabeled or partially-labeled datasets, comparison of feature representations, and post-hoc inspection and refinement of existing labels with interactive visual feedback. By combining multiple complementary 2D visualizations derived from high-dimensional feature representations, TSExplorer enables users to reason about complex temporal data without requiring deep expertise in dimensionality reduction or visualization engineering.

TSExplorer operates by projecting high-dimensional time-series feature representations into multiple complementary 2D visualizations. These projections allow users to inspect the structure of the dataset, identify clusters, outliers, and ambiguous regions, and interactively assign or refine labels. The tool is cross-platform, meaning it can be deployed across different operating systems and environments, which is essential for collaborative research and reproducibility.

TSExplorer addresses a practical need in time-series research: the ability to interactively explore and annotate data in a flexible, general-purpose manner. By decoupling feature representation from visualization and annotation, the tool allows researchers to plug in domain-specific feature extractors and projection methods while benefiting from a unified interactive interface. This design supports iterative workflows where labels are refined based on visual feedback, which is particularly valuable in scenarios with noisy or ambiguous data. The cross-platform nature of the tool enhances accessibility and collaboration. However, the source material does not provide details on the specific projection algorithms implemented, the user interface design, performance characteristics, or validation studies. Future work could include empirical evaluations of annotation efficiency and label quality, as well as integration with active learning loops. Overall, TSExplorer represents a contribution to the tooling ecosystem for time-series data, lowering the barrier to interactive data exploration and annotation.

CS practitioners and researchers

Opening member content…