PaperlayerView exact revision on arXiv

Source-linked AI summary

Spatiotemporal Data Mining: A Survey on Challenges and Open Problems

Ali Hamdi, Khaled Shaban, Abdelkarim Erradi, Amr Mohamed, Shakila Khan Rumi, Flora Salim

arXiv:2103.17128v1cs.LGcs.DB

TL;DR

STDM is difficult because it must capture complex spatiotemporal relationships across interdisciplinary data with distinctive characteristics. This survey synthesizes challenges and open problems across representations, modelling, tasks, and applications, highlighting unresolved issues in data integration and privacy.

  • Problem

    STDM requires methods for discovering patterns in large geographic and time-stamped datasets despite complex relationships, interdisciplinary tasks, and distinctive data characteristics.

  • Method

    The paper provides a literature survey of STDM challenges, limitations, and open research problems across data representations, modelling, visualisation, mining tasks, and applications.

  • Results

    The survey identifies unresolved issues spanning spatiotemporal representation, temporal modelling, pattern and outlier mining, cloud processing, and application-specific analyses.

  • Takeaways & Limitations

    Future STDM research must better represent spatiotemporal events and relationships and integrate spatial and temporal information in modelling and applications.

  • Takeaways & Limitations

    Sensor data can be sparse and difficult to integrate because different sources use different sampling rates, such as medical and GPS sensors.

Abstract

from arXiv · show

Spatiotemporal data mining (STDM) discovers useful patterns from the dynamic interplay between space and time. Several available surveys capture STDM advances and report a wealth of important progress in this field. However, STDM challenges and problems are not thoroughly discussed and presented in articles of their own. We attempt to fill this gap by providing a comprehensive literature survey on state-of-the-art advances in STDM. We describe the challenging issues and their causes and open gaps of multiple STDM directions and aspects. Specifically, we investigate the challenging issues in regards to spatiotemporal relationships, interdisciplinarity, discretisation, and data characteristics. Moreover, we discuss the limitations in the literature and open research problems related to spatiotemporal data representations, modelling and visualisation, and comprehensiveness of approaches. We explain issues related to STDM tasks of classification, clustering, hotspot detection, association and pattern mining, outlier detection, visualisation, visual analytics, and computer vision tasks. We also highlight STDM issues related to multiple applications including crime and public safety, traffic and transportation, earth and environment monitoring, epidemiology, social media, and Internet of Things.

1 Introduction

STDM has expanded with the growth of geo-referenced and temporal data, but existing surveys did not comprehensively organize its challenges and open problems. This survey addresses that gap across data types, relationships, research limitations, tasks, and applications.

  • Motivation: Growing geo-referenced and temporal data, including remote sensing, mobility, wearable-device, and social-media observations, has increased interest in STDM.These data span phenomena from cellular-scale evolution to climate change and support analysis of human needs and sentiments.
  • Data foundations: Spatiotemporal data combine non-spatiotemporal, spatial, and temporal attributes describing objects or fields across locations and time.The paper distinguishes events, trajectories, point references, and raster data as representative spatiotemporal forms.
  • Research gap: Existing surveys examined selected STDM perspectives, but did not thoroughly present STDM challenges and problems as a dedicated topic.Prior work covered spatial databases, patterns, clustering, urban computing, big-data analytics, climate data, outliers, and deep learning.
  • Survey scope: The survey reviews general challenges, then task- and application-specific challenges, and finally connects these perspectives through a mapping and discussion.Its organization includes methodology, general challenges, tasks, applications, integration, and conclusions.

2 Survey Methodology

The survey methodology builds a challenge-focused literature corpus by extending prior surveys with task and application coverage. It uses broad search and bibliographic screening across 342 works, emphasizing recent and highly ranked venues.

  • Challenge extraction: The authors construct a comprehensive challenge set from existing surveys and domain knowledge, then extend it to STDM tasks and applications.The approach also records root causes and organizes challenges through a proposed taxonomy.
  • Literature corpus: The review includes 342 STDM-related works from journals, conferences, books, book chapters, and theses.The corpus spans multiple publication types rather than a single venue category.
  • Venue coverage: Q1 journals account for 48 percent of the cited journal and conference distribution, followed by A* conferences at 29 percent.These ranks and quartiles were calculated using SJR and CORE classifications.

3 General STDM Challenges and Research Gaps

The survey attributes STDM difficulty to complex relationships, interdisciplinary integration, discretisation effects, dynamic and heterogeneous data, and incomplete research support. It presents these factors with their causes and extends them across tasks and applications.

  • Challenge organization: The survey identifies and lists the factors causing STDM difficulties before discussing each challenge and its causes in dedicated subsections.Figure 6 provides a cause-and-effect taxonomy of the general challenges.
  • General challenge factors: STDM challenges arise from complex implicit relationships, interdisciplinary data and algorithm integration, discretisation effects, heterogeneous dynamic data, and research gaps.The listed factors cover both intrinsic data properties and limitations in current STDM research.

3.1 Spatiotemporal Relationships

Spatiotemporal relationships are difficult to mine because they are complex, implicit, and non-identically distributed across space and time. Their dependencies and topological forms complicate pattern extraction and computation.

  • Relationship characteristics: Spatiotemporal relationships are harder to discover than conventional data relationships because they are complex, implicit, and non-identically distributed.The survey identifies these three characteristics as the central relationship challenges.
  • Complexity: Continuous real-world processes are represented through discrete spatial and temporal observations, complicating extraction of spatiotemporal patterns.Fixed traffic sensors, for example, sample continuously moving vehicles at selected locations.
  • Implicit relationships: Spatial relationships depend on qualities such as distance, volume, size, and time, and include topological relations such as overlap, disjoint, contains, and meet.These relations can connect points, lines, regions, or mixtures of object types.
  • Non-independent distributions: Positive spatial and temporal autocorrelation makes nearby objects more related than distant ones, while dependence can reduce algorithm performance and increase computational cost.Traffic persistence illustrates temporal dependence, and measuring autocorrelation in large datasets is computationally expensive.

3.2 Interdisciplinary and Combined Data Mining

STDM often requires combining datasets from multiple domains because phenomena such as crime and environmental change have interconnected spatial, temporal, and contextual drivers.

  • 3.2 Interdisciplinary and Combined Data Mining: Crime analysis combines macro-level socioeconomic, psychological, cultural, and demographic data with micro-environmental datasets.
  • 3.2 Interdisciplinary and Combined Data Mining: Figure 8 illustrates the need to integrate different data sources with crime datasets for interdisciplinary analysis.
  • 3.2 Interdisciplinary and Combined Data Mining: Combined mining addresses interdisciplinary relationships, including hybrid air-quality models using spatial, temporal, and inflection predictors.

3.3 Spatiotemporal Region Discretization

Spatiotemporal discretisation aggregates observations to make analysis feasible, but spatial scale and zoning choices can change the patterns and statistics obtained.

  • 3.3 Spatiotemporal Region Discretization: Discretisation aggregates spatiotemporal data to summarise information and extract features over ranges rather than single points.
  • 3.3 Spatiotemporal Region Discretization: Crime rates require aggregation across areal units because they cannot be measured meaningfully at a single spatial point.
  • 3.3 Spatiotemporal Region Discretization: Figure 9 depicts how scaling and zoning choices affect analysis results between alternative spatial units.
  • 3.3 Spatiotemporal Region Discretization: Scale effects alter statistical measures across aggregation levels, while zoning effects arise when areal-unit borders change.

3.4 Data Characteristics

STDM data characteristics create challenges because similarities can be semantically ambiguous, distributions evolve over time, and observations span social, networked, and representation-specific settings.

  • 3.4 Data Characteristics: STDM data exhibit specificity, vagueness, dynamicity, social and networking properties, heterogeneity, privacy concerns, and poor quality.
  • 3.4 Data Characteristics: Spatiotemporal models are domain-specific and may not generalise across applications or geographical areas with different characteristics.
  • 3.4 Data Characteristics: Spatially similar trajectories can be semantically different when contextual information, such as departure locations, is considered.
  • 3.4 Data Characteristics: Changing distributions and densities require dynamical models that capture moving-object evolution across timestamps.
  • 3.4 Data Characteristics: Social datasets combine behavioural text, images, videos, socioeconomic characteristics, and geo-temporal sensor data for pattern and trend analysis.
  • 3.4 Data Characteristics: Matching trajectories to road networks converts movement into edge-id sequences, making some trajectory mining and clustering tasks more straightforward.

3.4.6 Heterogeneity and Non-stationary

Heterogeneity, non-stationarity, privacy, data quality, scale, and infrastructure limitations constrain STDM across changing environments and large datasets.

  • 3.4.6 Heterogeneity and Non-stationary: Spatial heterogeneity and temporal non-stationarity create regional differences that can produce inconsistencies between global and regional models.
  • 3.4.6 Heterogeneity and Non-stationary: Privacy concerns arise because trajectories and mobile-call data can reveal personal movements and behavioural patterns.
  • 3.4.6 Heterogeneity and Non-stationary: Privacy-preserving STDM uses identity suppression, perturbation, sanitisation, and related modifications to protect personal and corporate data.
  • 3.4.6 Heterogeneity and Non-stationary: Data quality directly affects analysis results, while interdisciplinary datasets may be fragmented or distorted by uncertainty, partial knowledge, and conjecture.
  • 3.4.6 Heterogeneity and Non-stationary: Large-scale generation creates volume, variety, and velocity challenges for storing and processing spatiotemporal data.
  • 3.4.6 Heterogeneity and Non-stationary: Cloud frameworks support STDM management and analytics but remain limited in data sharing, scalability, interactive performance, and GIS visualisation.

3.5 Open problems in STDM Research

STDM remains challenged by methods that inadequately represent, model, and visualise the joint dynamics of space and time. Open problems also include integrating heterogeneous domains and developing comprehensive approaches rather than isolated task-specific solutions.

  • Classical data-mining techniques often perform poorly because STDM must discover relationships among events ordered across spatial and temporal dimensions.
  • Data representations: Spatiotemporal representation methods remain insufficient for capturing events and relationships, limiting subsequent STDM modelling.
  • Advanced modelling: Models that ignore temporal attributes or combine location and timestamps inadequately can produce poor hotspot detection and pattern-prediction outcomes.
  • Advanced modelling: Long-range temporal dependencies and occlusion remain difficult for spatiotemporal modelling, motivating approaches such as Transformers and GANs.
  • Visualisation: Spatiotemporal visualisation needs methods tailored to dynamic data because existing GIS and research largely prioritise spatial visualisation.
  • Comprehensiveness: Comprehensive STDM solutions must integrate multiple tasks, data types, and domains rather than addressing isolated problems independently.

4 STDM Task-related Challenges

STDM tasks face distinct challenges arising from spatiotemporal structure, including inadequate temporal modelling, difficult pattern search, weak evaluation, and the need to jointly visualise space and time.

  • 4.1 Spatiotemporal Prediction: Spatiotemporal prediction uses features such as speed, acceleration, duration, distance, length, and direction with classifiers, regressors, ensembles, or deep learning models.
  • 4.1 Spatiotemporal Prediction: LIME explains individual model predictions by fitting an interpretable local model around the prediction and identifying contributing symptoms.
  • 4.2 Spatiotemporal Clustering: Clustering evaluates similarity across multiple spaces, making conventional Euclidean-space evaluation insufficient for spatiotemporal objects and trajectories.
  • 4.3 Spatiotemporal Pattern Mining: Spatiotemporal pattern mining faces exponentially many possible patterns, absent explicit transactions, and potential over-counting, creating accuracy–efficiency trade-offs.
  • 4.4 Spatiotemporal Outlier Detection: Outlier detection is constrained by model generalisation, scalability, limited spatiotemporal representations, and low interpretability.
  • 4.6 Spatiotemporal Visual Analytics: Visual analytics must represent spatial and temporal information simultaneously, while existing methods often emphasise global origin–destination patterns over local stops and segments.

5 STDM Application-related Challenges

STDM applications must handle domain-specific combinations of large, heterogeneous, dynamic, and interconnected data. Challenges span public safety, transport, environmental monitoring, epidemiology, social media, and IoT.

  • 5.1 Crime and Public Safety: Crime analysis lacks comprehensive methods for heterogeneous historical, geographical, and demographic data and systematic representation of temporal crime attributes.
  • 5.2 Traffic and Transportation: Traffic applications must collect, store, and process large dynamic datasets while combining transportation, accident, injury, road-network, and criminal-record information.
  • 5.2 Traffic and Transportation: Ride-hailing fairness is difficult because matching strategies can distribute jobs unfairly among drivers, motivating analysis of matching distributions over time.
  • 5.4 Epidemiology and Spread of Infectious Diseases: Epidemiological mining must account for high-volume mobility across timescales, evolving pathogens, cross-border spread, and interactions among geographic areas and health systems.
  • 5.5 Social Media Analysis: Social-media analysis must model complex spatiotemporal correlations in check-ins, reviews, and geo-temporally tagged posts.
  • 5.6 Smart Internet of Things: IoT data integration is complicated by sparsity, different sensor sampling rates, measurement veracity, and uncertainty in spatial localisation and temporal synchronisation.

6 Summary of STDM General Challenges

The survey identifies general STDM challenges arising from complex relationships, interdisciplinary and combined data mining, region discretisation, heterogeneous and dynamic data, and limited representations.

  • Spatiotemporal relationships: STDM relationships involve implicit dependencies, autocorrelation, non-identical distributions, and influence between co-located objects or trajectories.These characteristics include exponential relationship counts and wide variation in data distributions across space and time.
  • Interdisciplinary and Combined Data Mining: Interdisciplinary STDM combines heterogeneous datasets, multiple domains, and multiple mining techniques, increasing integration complexity.The survey links this challenge to combined data mining across related domains and datasets.
  • Region Discretisation: Region discretisation introduces scale and zoning effects that make mining results dependent on the selected spatial regions.The summary table associates region discretisation with scale effect and scale dependency across STDM tasks.
  • Data characteristics: Spatiotemporal data are heterogeneous, dynamic, non-stationary, and often unique to particular space-time regions.These properties require different learning models for varying regions and complicate generalisation across space and time.
  • Research limitations: STDM still needs improved data representations because existing representations may depend on high-density locations while omitting temporally related attributes.The survey identifies limited representations as a research problem alongside privacy, uncertainty, partial knowledge, and big-data constraints.

7 Conclusion

The paper surveys STDM challenges across general issues, tasks, and applications, then identifies open gaps in representations, modelling, visualisation, and integrated approaches. It concludes that future work should develop methods integrating multiple STDM tasks for more complex scenarios.

  • The survey covers general STDM issues involving relationships, interdisciplinarity, discretisation, data characteristics, and research limitations.
  • Future research should develop modelling and visualisation methods that integrate multiple STDM tasks for complex scenarios.
  • It extends the review to challenges associated with STDM tasks and applications.
Loading 2103.17128v1…