PaperlayerView exact revision on arXiv

Source-linked AI summary

Time Series Information Visualization -- A Review of Approaches and Tools

Evandro S. Ortigossa, Fábio F. Dias, Diego C. Nascimento, Luis Gustavo Nonato

arXiv:2507.14920v2cs.GR

TL;DR

Large, multidimensional time-series datasets are difficult to represent and analyze clearly, especially when many features evolve over time. This review synthesizes visualization approaches, interaction strategies, theoretical foundations, and design guidelines for time-series InfoVis. It identifies current tools and generative visualization developments while highlighting persistent challenges in pattern tracking, coordination, scalability, and real-time exploration.

  • Problem

    Large time-series datasets with multiple evolving features create representation, cognitive-load, and scalability challenges for visual analysis.

  • Method

    The paper reviews time-series InfoVis approaches, tools, interaction strategies, theoretical concepts, graphical properties, and design guidelines, emphasizing multidimensional data.

  • Results

    The review identifies diverse context-dependent visualization strategies and highlights modern tools, including generative visualization models, for improving visual representation.

  • Takeaways & Limitations

    Developing clear, comprehensible, interactive, and computationally scalable visualization systems remains fundamental for analyzing increasingly complex time-series data.

  • Takeaways & Limitations

    Some techniques cannot systematically track information or detect patterns because they primarily visualize relationships between data records.

Abstract

from arXiv · show

Time series data are prevalent across various domains and often encompass large datasets containing multiple time-dependent features in each sample. Exploring time-varying data is critical for data science practitioners aiming to understand dynamic behaviors and discover periodic patterns and trends. However, the analysis of such data often requires sophisticated procedures and tools. Information visualization is a communication channel that leverages human perceptual abilities to transform abstract data into visual representations. Visualization techniques have been successfully applied in the context of time series to enhance interpretability by graphically representing the temporal evolution of data. The challenge for information visualization developers lies in integrating a wide range of analytical tools into rich visualization systems that can summarize complex datasets while clearly describing the impacts of the temporal component. Such systems enable data scientists to turn raw data into understandable and potentially useful knowledge. This review examines techniques and approaches designed for handling time series data, guiding users through knowledge discovery processes based on visual analysis. We also provide readers with theoretical insights and design guidelines for considering when developing comprehensive information visualization approaches for time series, with a particular focus on time series with multiple features. As a result, we highlight the challenges and future research directions to address open questions in the visualization of time-dependent data.

1 Introduction

Time series visualization is needed to make large, multidimensional, time-varying datasets interpretable despite representation and scalability challenges. This review synthesizes approaches, design principles, interaction strategies, and tools for multidimensional time-series analysis.

  • Multiple correlated or uncorrelated attributes changing over time make time-series representation difficult, even in systems designed for multidimensional data.
  • Visualization reduces the cognitive effort required to identify and interpret patterns by transforming abstract data into graphical representations.
  • Line-based charts often become cluttered and occluded when representing multiple series, motivating advanced approaches across application domains.
  • The review examines conceptual, practical, and aesthetic considerations for developing comprehensive tools, particularly for multidimensional time-series analysis.
  • Its contributions include reviewing graphical principles, visualization challenges, interactive tools, and applications for time-series InfoVis.

2 Method of the Systematic Review

The review applies a systematic literature-search and screening process to identify research on time-series information visualization. It combines multiple scholarly databases with explicit inclusion criteria and full-text review.

  • The search combined IEEE Xplore, ACM Digital Library, DBLP, Scopus, Google Scholar, ScienceDirect, and Web of Science.
  • Queries targeted multidimensional time series and InfoVis in titles, abstracts, and keywords, covering mainly 2005–2024 while including seminal studies.
  • Eligible studies addressed time-series representation and were drawn primarily from peer-reviewed visualization venues, with additional preprints, theses, and books.
  • After duplicate removal and screening, 232 studies were read in full and 219 references were reported.

3 Previous Visualization Reviews

Earlier reviews established broad taxonomies and methods for visualization, multidimensional data, temporal analysis, interaction, and scalability. This review focuses specifically on updated interactive techniques and tools for multidimensional time-series InfoVis.

  • Prior surveys covered general visualization, visual data mining, multifaceted scientific data, interactive exploration, and multidimensional visualization.
  • Reviews of large-data visualization emphasized working-memory limits, scalability, focus+context exploration, and selective presentation of information.
  • Temporal visualization surveys categorized time-oriented data and interactive techniques according to temporal structure and dynamic behavior.
  • Despite this literature, effectively visualizing multidimensional time series remains challenging and requires focused coverage of interactive solutions.
  • The review extends previous work by discussing techniques and tools that were previously unavailable or only recently emerged.

4 Ground Theory

The ground theory defines time-series data, their multidimensional structure, and the representational challenges created by increasing temporal and feature complexity. These concepts motivate data-cube abstractions and specialized visualization tools.

  • 4.1 Time Series Notation and Nomenclature: Time-series data consist of time-oriented observations whose values may be numerical, categorical, images, binary sequences, or other feature types.
  • 4.1 Time Series Notation and Nomenclature: A multidimensional time series can be represented conceptually as a data cube with samples varying across time slices and features, extendable with spatial dimensions and frequency bands.
  • 4.2 Multidimensionality: A multidimensional time series contains multiple features per time-dependent instance, whether or not those features are dependent at each time interval.
  • 4.2 Multidimensionality: Growing numbers of instances and attributes make time-series analysis complex and difficult to interpret.
  • 4.2 Multidimensionality: As time-cube dimensionality increases, representing and interpreting its content becomes harder, complicating extraction of trends and interdependencies among time-varying features.

4.3 Fragmented and Intermittent Time Series

Fragmented time series arise from noise, sensor failures, communication loss, and irregular sampling, requiring methods that address missing observations without obscuring temporal dynamics.

  • Missing observations can obscure underlying patterns and bias inference, so fragmented series require explicit treatment that preserves temporal dynamics.
  • Classical interpolation, ARIMA, Kalman filters, and state-space methods estimate short missing spans when dynamics are locally smooth.These approaches are presented as one broad category for handling fragmented series.
  • Visualization should expose missingness and provenance while pairing imputed values with uncertainty encodings such as error bars or translucent bands.Interactive controls can let analysts switch between raw and imputed views.

4.4 Overlook Through Time Series Modeling

Time-series modeling supports visualization in both time and frequency domains by decomposing observations into interpretable components or spectral patterns.

  • Model outputs and filter states, including time-varying parameters and phase evolutions, can also be visualized beyond raw observational dynamics.
  • Time-domain models decompose series into components such as trend, seasonality, cycles, and stochastic volatility, combined additively or multiplicatively.
  • Classical frameworks include (S)ARIMA, GARCH, Holt-Winters, SEATS, X-11, STL, ETS, dynamic regression, TBATS, hidden Markov models, harmonic regression, and mSSA.
  • Modern machine-learning approaches include EfficientNet-based [1], Temporal Fusion Transformer, and Vision Transformer-based [141] models.
  • Frequency-domain representations combine sinusoidal components across frequencies, while Fourier transforms such as DFT and DCT detect periodicities and frequency content.
  • Wavelet transforms represent non-stationary series through localized time-frequency components, capturing energy transfer and phase changes that Fourier transforms can overlook.
  • Coordinated views can visualize time-domain components dynamically, while spectral plots emphasize dominant periods and recent models integrate spectral coefficients with ANN models [77] [211].

4.5 Time Series and Information Visualization

Information visualization combines data-processing stages, graphical representations, and interaction to support pattern discovery and knowledge extraction from large time-series datasets.

  • Visual mining combines data-mining and InfoVis techniques by embedding interactive visual strategies into the knowledge-discovery process.
  • The visualization pipeline acquires and organizes raw data, preprocesses and transforms attributes, maps them into graphical representations, and supports interactive exploration.
  • Graphical algorithms transform numerical and categorical attributes into adjustable visual formats that capture temporal evolution.
  • Interaction lets analysts select intervals, compare views, filter noise, and convert visualized information into actionable knowledge.
  • Connecting analysts with an interactive information space supports cognitive knowledge discovery and perception of previously unknown patterns.
  • Minard’s chart integrates spatial, temporal, and quantitative variables, including troop numbers and temperature, within one multidimensional representation.

4.6 Time Series InfoVis Taxonomies and Guidelines

Time-series InfoVis taxonomies organize representations by spatial layout and interaction, while guidelines emphasize coordinated views that manage dimensionality without sacrificing comparison or interpretability.

  • Taxonomies and design properties: Taxonomies classify time-series visualizations by properties such as geometry, iconography, information access, layout, and shared- versus split-space organization.
  • Shared-space and split-space approaches: Shared-space visualizations plot multiple series together for direct comparison, but more than four curves can cause clutter, cognitive overload, and reduced readability.
  • Shared-space and split-space approaches: Split-space techniques assign each series its own small-multiple chart, reducing overload through comparable side-by-side layouts.
  • Shared-space and split-space approaches: Figure 3 contrasts a shared-space stream graph, with stacked layers along a central time axis, against split-space small area charts for U.S. unemployment rates from 2000–2010.
  • Guidelines for coordinated views: Spatial placement should minimize context switching, while interconnected views support interactive browsing and manageable divide-and-conquer exploration.
  • Guidelines for coordinated views: Hybrid interfaces require synchronized interactions across views, making coordination difficult and allowing coding errors to propagate across interconnected components.
  • Guidelines for coordinated views: Multi-view systems divide complex information into linked views, enabling comparisons from multiple perspectives and reducing working-memory demands.

4.7 Graphical Perception and Visual Properties

Graphical perception principles help visualization systems represent multidimensional time series while limiting clutter, occlusion, and cognitive overload. The review emphasizes scalable visual properties, perceptual grouping, and selective exploration as complementary design requirements.

  • Visual scalability: Visual scalability is constrained by human working memory, while overlapping graphical symbols can obstruct information and reduce interpretability.Increasing curve counts can also decrease discriminability, making individual series harder to inspect.
  • Visual scalability: Increasing time-varying features from three to ten reduces individual legibility in line charts, producing visual disorder and clutter.The example uses unemployment-rate data across U.S. industrial sectors from 2000 to 2010.
  • Perceptual grouping: Gestalt grouping principles organize elements through shared characteristics including proximity, similarity, common action, continuity, and enclosed boundaries.These principles support the construction of more complex perceptual groupings from graphical elements.
  • Graphical properties: Graphical properties such as spatial placement, size, orientation, shape, and color influence perceived importance and the distinction between variables.High-contrast colors facilitate accurate reading, whereas insufficient contrast increases the effort needed to individualize adjacent objects.
  • Exploratory support: Because excessive simultaneous information overloads users, visualization systems should support selecting and reorganizing content at different levels of granularity.The review also discusses focus on suitable colormaps, noting that context determines whether rainbow-based designs are appropriate.

4.8 Interactive Exploratory Techniques

Interactive exploratory techniques connect visual representations with user actions to manage dense multidimensional time series. The review covers coordinated views, focus and detail mechanisms, responsive transitions, and device-specific interaction constraints.

  • Coordinated views: Dense datasets and page scrolling can fragment users’ overview, so multi-view systems link selections across charts and provide multiple abstraction levels.Linked highlighting helps users coordinate information that cannot be viewed simultaneously in a single display.
  • Overview and detail: Interactive systems reduce visual complexity by combining overview, zoom, filtering, and details-on-demand with coordinated views.Selecting items or intervals can update linked views to reveal detailed information without changing the overall representational context.
  • Focus and detail: Focus+context and details-on-demand use magnification or zooming to expose focal information while preserving contextual or spatial reference.Distortion lenses can magnify dense regions locally, while zooming relies on users’ spatial memory to compare information across scales.
  • Responsive interaction: Fluid, immediate responses are important because delays between interactions and view updates can disrupt users’ expectations and exploratory mental models.The review presents responsiveness and coordinated transitions as design considerations for interactive visualization tools.
  • Responsive interaction: Animation can improve perception of object changes across correlated charts, but complex transitions may induce misinterpretation and should typically respond within one to three seconds.Heer and Robertson recommend simple transitions that support screen updates without violating visual semantics.
  • Mobile interaction: Touch-screen visualization must accommodate tablets and smartphones’ smaller displays and interaction constraints through gestures such as pinching to zoom and dragging to select.Tablets and smartphones generally provide less screen space than regular computers, with screens ranging roughly from four to thirteen inches.

4.9 Large-Scale and Real-Time Time Series

Large-scale time series require methods that preserve computational efficiency, perceptual clarity, and interactivity as data volumes grow. The review highlights aggregation and downsampling, multi-view refinement, symbolic representations, and progressive processing for real-time analysis.

  • Scalability strategies: Extreme-scale time series with millions or billions of samples require aggregation or downsampling alongside multi-view systems that progressively refine detail.These complementary strategies reduce the visual and computational burden while preserving access to relevant information.
  • Aggregation and downsampling: Temporal binning groups samples into fixed intervals and summarizes them with statistics such as averages, medians, or densities to preserve high-level trends.The review cites imMens for subsecond querying through binned aggregation and TiVy [42] for adaptive visual summaries.
  • Aggregation and downsampling: Curve-preserving downsampling selects representative points to retain salient visual features while reducing the number of displayed samples.This approach complements aggregation when preserving the shape of time-series curves is important.
  • Symbolic representations: Symbolic Data Analysis can compress multidimensional time series while revealing temporal patterns beyond simple average trends.Nascimento et al. summarized more than 1.5 billion 256-channel EEG observations into 15 million symbolic intervals, approximately 0.98% of the original size.
  • Real-time processing: Real-time systems use incremental aggregation and pattern extraction to process data in near-constant time, with deep learning approaches supporting clustered visualizations through embeddings.Progressive partitioning updates binned summaries on the fly, while neural methods integrate visual analytics with learned representations.

5 Visual Approaches and Methods

The review organizes time-series visualization approaches by visual metaphor, emphasizing their strengths, limitations, and suitability for increasingly large and multidimensional data. Examples include spiral, calendar, heatmap, and coordinated map-based views for revealing periodic, temporal, and spatial patterns.

  • Visual approaches are difficult to categorize as time-varying datasets grow in size and dimensionality, prompting increasingly sophisticated solutions that combine conceptual frameworks.
  • The review groups time-series InfoVis techniques by related visual metaphors while comparing each approach’s strengths and limitations.
  • 5.1 Spiral and Radial Charts: Spiral charts reveal overlapping cycles across granularities, but require careful parameterization and offer limited advantages for long periodic cycles.A Seattle ringmap covering 2007–2016 highlighted seasonal low temperatures from December through March.
  • 5.2 Calendar View and Heatmaps: Calendar and heatmap views provide familiar, compact representations of fine-grained temporal data, but excessive density can cause clutter and hinder interpretation.Clustering, hierarchical methods, temporal compression, interaction, or complementary visual metaphors can mitigate these limitations.
  • 5.3 Time, Space, and Maps: Flowstrates coordinate origin and destination maps with a central heatmap, enabling separate exploration of spatial and temporal flow patterns.In refugee-movement case studies, Somalia historically had the strongest destination flow, but occlusion and inaccurate travel-distance perception remained limitations.

6 Applications

The applications section reviews composed and domain-specific systems that combine coordinated visual and analytical techniques for time-varying data. These systems expose evolving topic, graph, financial, and transaction patterns while retaining challenges involving clutter, occlusion, computational cost, and cognitive load.

  • 6 Applications: Composed interfaces integrate multiple coordinated techniques to guide interpretation and mitigate limitations of single-method visualizations.
  • 6.1 Textual Documents and Topics: EvoRiver separates positive and negative topic behaviors into opposing stream layers, with interactive links and keyword views supporting analysis of changing topic relationships.During an election analysis, “Government” and “Politics” alternated between competitors and collaborators under the influence of “International issues”; sentiment classification may introduce subjective bias.
  • 6.2 Graph-structured Time Series: Graph-structured time-series systems represent evolving nodes, links, clusters, or transitions, but large datasets remain difficult to render because of computational demands and occlusion.Clustering, hierarchical aggregation, efficient layouts, zooming, fisheye distortion, and color encodings are used to support navigation and reduce overlap.
  • 6.3 Finance and Business: Financial and transaction applications combine clustering, heatmaps, graph-like views, word networks, and machine learning to inspect market behavior, fraud, anomalies, and trends.Coordinating multiple synchronized views can increase cognitive and perceptual load, whereas automated anomaly detection can reduce context switching.

7 Discussions and Future Research on Time Series InfoVis

Time-series InfoVis faces persistent scalability, dimensionality, and real-time responsiveness challenges, while current tools largely extend traditional visual metaphors with interaction and coordinated analytical views. The review identifies domain-sensitive design, user interaction, and integrated tooling as central directions for handling complex multidimensional data.

  • Discussions and Future Research: Modern tools increasingly combine graphical metaphors with interaction mechanisms to support exploration of multidimensional time series and reduce limitations imposed by high dimensionality.The review emphasizes interactive-based tools, methods, and approaches designed specifically for multidimensional time series.
  • Tools and Applications: Available implementation options span open-source libraries, interactive web technologies, business-intelligence platforms, and generative models, each with different capabilities and constraints.Table 4 summarizes tools and libraries, while proprietary platforms may require expensive licensing.
  • Challenges and Open Questions: Multidimensional time series require simultaneous observation of multiple features, making conventional line-based visualizations vulnerable to clutter and occlusion.The review also identifies the curse of dimensionality and big-data scale as major contemporary challenges.
  • Challenges and Open Questions: Real-time visualization remains difficult because large datasets increase processing and rendering times while dynamic information may require periodic updates.Robust systems must remain computationally scalable and responsive when handling large amounts of data.
  • Challenges and Open Questions: Anomaly detection remains challenging because anomalous behavior is context-dependent and unknown patterns are difficult to define precisely.The review connects visual exploration and interaction with pattern and anomaly discovery, while also noting Bayesian time-series clustering as a promising related approach.
  • Future Directions: Effective visualization design depends on application context, domain experts’ requirements, graphical properties, cognitive capacities, and the quality and dimensions of display devices.The review also highlights coordinated tools for data handling, processing, and presentation as a promising direction for large and complex datasets.

8 Conclusion

The conclusion presents visualization as fundamental for making temporal dependencies in large, multidimensional datasets perceptually intuitive. It emphasizes that effective systems depend on appropriate graphical metaphors, interaction, user-centered design, rendering conditions, and increasingly generative visualization models.

  • Conclusion: Well-designed visualization can transform abstract, time-oriented phenomena into perceptually intuitive visual elements that help analysts uncover hidden information.
  • Conclusion: Visualization effectiveness depends on graphical metaphors, interactive manipulation, user-centric cognitive alignment, and device quality and dimensions.The review synthesizes these concepts as ways to mitigate challenges in visual data exploration.
  • Conclusion: Generative visualization models have achieved prominent results in improving the understanding, synthesis, and optimization of visual representations.The conclusion frames clear and comprehensible layouts as fundamental for extracting meaningful information from increasingly complex data.
Loading 2507.14920v2…