PaperlayerView exact revision on arXiv

Source-linked AI summary

Analytics for the Internet of Things: A Survey

Eugene Siow, Thanassis Tiropanis, Wendy Hall

arXiv:1807.00971v1cs.DB

TL;DR

IoT deployment produces growing volumes of data, but the literature has not deeply explored their analytics across applications and infrastructure. The paper systematically reviews IoT analytics, proposes a capability-based taxonomy, and surveys enabling infrastructure. It concludes that analytics already plays a role across many IoT domains and that its importance will grow as infrastructure and device deployment scale.

  • Problem

    IoT data are growing in scale and pose volume, velocity, and variety challenges, while the IoT-analytics intersection has not been explored in depth.

    Source: §1 · pp. 1, 4

  • Method

    The authors systematically review IoT analytics applications, develop a capability-based classification and layered taxonomy, and survey enabling infrastructure.

    Source: §1 · p. 4, §4 · p. 7, §5 · p. 12

  • Results

    Analytics plays a role in applications across many IoT domains, and the review produces a taxonomy categorizing analytics by capabilities and application potential.

    Source: §5 · p. 12, §8 · p. 26

  • Takeaways & Limitations

    The taxonomy is intended to inform research and application roadmaps as IoT infrastructure develops and device deployment becomes ubiquitous.

    Source: §8 · p. 26

    The Lambda architecture, Hadoop, and Spark suit centralized or cloud clusters rather than decentralized fog- and edge-based computing.

    Source: §6.4 · p. 20

Abstract

from arXiv · show

The Internet of Things (IoT) envisions a world-wide, interconnected network of smart physical entities. These physical entities generate a large amount of data in operation and as the IoT gains momentum in terms of deployment, the combined scale of those data seems destined to continue to grow. Increasingly, applications for the IoT involve analytics. Data analytics is the process of deriving knowledge from data, generating value like actionable insights from them. This article reviews work in the IoT and big data analytics from the perspective of their utility in creating efficient, effective and innovative applications and services for a wide spectrum of domains. We review the broad vision for the IoT as it is shaped in various communities, examine the application of data analytics across IoT domains, provide a categorisation of analytic approaches and propose a layered taxonomy from IoT data to analytics. This taxonomy provides us with insights on the appropriateness of analytical techniques, which in turn shapes a survey of enabling technology and infrastructure for IoT analytics. Finally, we look at some tradeoffs for analytics in the IoT that can shape future research.

1 INTRODUCTION

IoT deployment and data production are growing, while IoT data bring big-data challenges of volume, velocity, and variety. Existing surveys address related topics, but the authors identify an underexplored focus on IoT analytics and set out to review applications, classify analytics, and examine infrastructure and tradeoffs.

1 INTRODUCTION

  • Connected IoT devices have grown rapidly, and the introduction presents big data as both a source and a sink of IoT activity.It reports a 300% increase in connected devices over five years in a 2013 report and gives projected annual economic-impact estimates from reports published in 2013 and 2015.Source: §1 · p. 1
  • IoT data pose volume, velocity, and variety challenges, from very large event streams to fast sensor data and heterogeneous sources.The introduction illustrates volume with 4 billion events from 40 homes in a month and notes that 71% of surveyed data scientists found analytics increasingly difficult because of data variety.Source: §1 · p. 1
  • Earlier reviews cover IoT, big data, analytics, and enabling technologies, but the authors say their intersection focused on IoT analytics has not been explored in depth.The paper positions its review against prior surveys of IoT applications, infrastructure, and analytics.Source: §1 · pp. 3–4
  • The paper reviews IoT analytics applications, proposes a classification and taxonomy, and surveys enabling infrastructure for big-data use cases.It also examines tradeoffs intended to shape future research directions.Source: §1 · p. 4

2 METHODOLOGY AND ORGANISATION OF ARTICLE

The authors use an evidence-based systematic review to survey IoT analytics applications and enabling infrastructure. They search and screen publications, synthesize an application sample, and review infrastructure across the data lifecycle.

2 METHODOLOGY AND ORGANISATION OF ARTICLE

  • The review frames two questions: which IoT analytics research and applications are published, and what infrastructure enables big-data IoT analytics.These questions guide the application survey and infrastructure review.Source: §2 · p. 4
  • A Web of Science search for “big data” or “analytics” filtered by “internet of things” retrieved 460 articles from 2011–2015 and 311 more from 2016–2017.The later set was added when the paper was revised.Source: §2 · p. 4
  • Manual screening selected 30 original-research papers with designs, implementations, results, analytics, and IoT use cases: six top-ranked papers from each of five application domains.Ranking combined citation counts and a 0–5 qualitative score for technological complexity and completeness to mitigate recency bias.Source: §2 · pp. 4–5
  • The authors use business analytics literature to develop an IoT analytics classification and taxonomy, then inspect cloud, middleware, distributed, and fog-computing work for infrastructure.The infrastructure review considers data generation, collection, integration, storage, and compute.Source: §2 · p. 5

3 THE INTERNET OF THINGS VISION AND APPLICATION DOMAINS

The paper presents the IoT as a globally interconnected system of identifiable sensing or actuating things that share data and may support socially impactful services. It organizes advanced-service application areas by their impacts and groups them into domains to structure the survey.

3.1 IoT Definition and Common Vision

  • The IoT vision combines global scale, identifiable physical things with sensing or actuating capabilities, interconnection for data sharing, and potential societal impact through advanced services.The paper grounds its motivation in shared and integrated IoT data as a basis for analytics and advanced services.Source: §3.1 · p. 5

3.2 IoT Advanced Services and Application Domains

  • Application areas elicited from prior surveys are classified by their impact on environment, society, and economy.These themes are presented as drivers of sustainable development and organize the application-area categorization.Source: §3.2 · p. 5, Figure 2 · p. 6
  • Health, transport, living, environment, and industry domains group application areas, while topics such as smart cities can span multiple impact themes.This classification is intended to organize and ensure broad coverage of the analytics and infrastructure survey.Source: §3.2 · pp. 5–7, Figure 3 · p. 6

4 IOT APPLICATIONS WITH ANALYTICS

The survey reviews analytics applications across IoT domains, showing how connected data support insight and action. Its examples span health, transport, living, environment, and industry, with predictive analytics prominent in industry and supply-chain applications.

4 IOT APPLICATIONS WITH ANALYTICS

  • Connected IoT data and context provide a wider observation space for smart parking, guiding users to spaces without human intervention and reducing traffic and pollution.The survey samples the top six papers in each application domain using its systematic-review methodology; Table 2 summarizes techniques, data sources, and whether data are historical or real-time.Source: §4 · p. 7

4.1 Health: Ambient Assisted Living, Neo-natal care, Prognosis, Monitoring

  • Health applications use contextual home data for advisories and anomaly warnings, wearable data for diagnosis and prediction, and multiple streams to identify neonatal conditions.Other examples include prognosis from comparable patients’ data and activity recognition that recommends activities.Source: §4.1 · p. 7

4.2 Transport: Traffic Control and Routing, Pedestrian Detection, Smart Parking

  • Transport analytics predicts traffic flow and parking availability, discovers patterns and anomalies, and supports route planning and pedestrian detection.Route recommendations rely on predicted costs and an admissible search heuristic; parking examples combine availability, location, traffic, and reservations.Source: §4.2 · p. 9

4.3 Living: Cultural Behaviour, Public Safety, Smart Buildings, Memory Augmentation, Lifestyle Monitoring

  • Living applications analyze movement, video, building sensors, and digital traces for cultural insights, public safety, anomaly diagnosis, memory augmentation, and activity monitoring.Building sensor definitions can support automatic derivation of diagnosis rules and behavior specific to a building.Source: §4.3 · pp. 9–10

4.4 Environment: Disaster Detection & Response, Wind Forecasting, Smart Energy

  • Environmental applications combine heterogeneous sensor and social-media data for disaster awareness, use wind-sensor histories for forecasting, and analyze smart-plug patterns to identify energy savings.Smart-energy savings included reduced electricity use from appliances, desktops, and printers during non-office hours.Source: §4.4 · p. 10

4.5 Industry: Supply Chain Management, Smart Farming, Chemical Process

  • Industry applications use real-time sensing and prediction for supply-chain decisions, smart farming, and process control; predictive analytics is common in product and supply-chain management.Large IoT datasets provide observation space for pattern discovery, while real-time sensor data can drive actuators to address out-of-stock products or unsuitable truck conditions.Source: §4.5 · p. 11

5 TYPES OF ANALYTICS AND THEIR IMPORTANCE

The paper classifies IoT analytics into five capabilities and relates them to the knowledge hierarchy. The framework distinguishes what each capability seeks to achieve and highlights its value.

5 TYPES OF ANALYTICS AND THEIR IMPORTANCE

  • The paper derives a comprehensive classification of IoT analytics with five capabilities: descriptive, diagnostic, discovery, predictive, and prescriptive.The classification builds on business-analytics literature; surveyed applications can support multiple capabilities, and all five are represented in the literature, while mature industrial IoT focuses on high-value analytics.Source: §5 · pp. 11–12
  • The five-category framework extends descriptive, predictive, and prescriptive analytics with diagnostic and discovery capabilities.The paper attributes the additional categories to Gartner and Corcoran et al..Source: §5 · p. 12
  • Fig. 4 relates analytical capabilities to the knowledge hierarchy, using the progression from data to information, knowledge, and wisdom to frame their aims and value.The survey connects this progression to organizational use of analytics, from hindsight to insight and prescribing actions.Source: §5 · pp. 12–13, Figure 4 · p. 12

5.1 Five Categories of Analytics Capabilities

The paper distinguishes five analytics capabilities by the questions they address, moving from describing and explaining past data to discovering insights, predicting outcomes, and recommending actions.

5.1.1 Descriptive Analytics.

  • Descriptive analytics answers “what happened?” by contextualizing, fusing, and presenting raw IoT data as charts, reports, statistics, or aggregates.It provides hindsight on observed data.Source: §5.1.1 · p. 13

5.1.2 Diagnostic Analytics.

  • Diagnostic analytics seeks root causes and explanations for IoT data, extending descriptive analytics’ account of what happened to why it happened.Both descriptive and diagnostic analytics provide hindsight.Source: §5.1.2 · p. 13

5.1.3 Discovery in Analytics.

  • Discovery analytics uses inference, reasoning, or detection to find novel trends, exceptions, or clusters not already known from raw IoT data.It can narrow the search space for analytics applications amid big-data volume.Source: §5.1.3 · p. 13

5.1.4 Predictive Analytics.

  • Predictive analytics uses past data and knowledge to estimate likely future outcomes and assess prediction quality.Source: §5.1.4 · p. 13

5.1.5 Prescriptive Analytics.

  • Prescriptive analytics uses foresight and uncertainty-aware optimization to recommend a timely course of action and evaluate “what if” alternatives.Source: §5.1.5 · p. 13

5.2 Specific Types of Analytics

The section classifies specific analytics approaches used with IoT data. It describes their roles, from interactive visual analysis to business decision support.

5.2.1 Visual Analytics.

  • Visual analytics combines interactive visualizations with data analytics for understanding, reasoning, and decision-making, and supports description, diagnosis, and discovery.The cited healthcare example uses it to explore pregnancy-age distributions, hypothesize disease patterns, and detect a delayed outbreak.Source: §5.2.1 · pp. 13–14

5.2.2 Data Mining.

  • Data mining includes OLAP summaries, association and correlation, classification, unlabeled clustering, pattern discovery, and anomaly detection.OLAP supports drill-down and roll-up; association and correlation examine relationships between variables.Source: §5.2.2 · p. 14

5.2.3 Content and Text Analytics.

  • Content analytics applies analytical techniques to digital content, while text analytics derives information from unstructured text, such as entities, relations, sentiment, events, and time-series information.Source: §5.2.3 · pp. 14–15

5.2.4 Video Analytics.

  • Video analytics uses specialized software and hardware to analyze captured footage and identify objects, events, behavior, or attitudes in real time.Source: §5.2.4 · p. 15

5.2.5 Trend Analytics.

  • Trend analytics examines data and events over time to understand patterns, predict future trends, and provide early warning systems.It is closely related to time-series analysis, including identifying long-term changes in the mean level.Source: §5.2.5 · p. 15

5.2.6 Business Analytics.

  • Business analytics uses organizational data to inform business decisions and automate or optimize business processes.Source: §5.2.6 · p. 15

5.3 A Layered Taxonomy of Data, Analytics and Applications for the IoT

The taxonomy organizes the survey into data, analytics, and application layers. It connects IoT applications in surveyed domains to analytics capabilities and techniques, alongside data characteristics and distributed infrastructure.

Analytics layer

  • The analytics layer covers capabilities and specific analytics, drawing on established literature for visual analytics, data mining, text and video analytics, forecasting, control, and business analytics.Source: §5.3 · p. 15

Application layer

  • The application layer connects surveyed IoT applications across domains to analytics capabilities, with techniques referenced under each capability.Source: §5.3 · p. 15

Data layer

  • The data layer summarizes big-data terms and sources, including veracity as data noise and accuracy and variability as context-dependent changes in meaning; it also covers storage and compute distribution and technologies.Source: §5.3 · p. 15

6 ENABLING INFRASTRUCTURE FOR IOT ANALYTICS

This section surveys infrastructure and technologies that enable IoT analytics across the data flow, from data generation and collection through integration, storage, and compute. It also classifies the distribution of storage and compute and provides a basis for future work on IoT big data analytics challenges.

6.1 Data Generation: Sensors and Tags, Hardware and OS, Power

  • IoT data generation relies on sensors and passive or active tags, while remotely deployed sensors need power for wireless transmission and may use energy harvesting or charging alternatives.The surveyed sensors include environmental, spatial, and health types; tags include QR codes, barcodes, iBeacon, and RFID.Source: §6.1 · p. 18

6.2 Data Collection: Discovery, Management, Transmission, Context and Fog

  • Collection infrastructure covers middleware functions, device discovery and lookup, data transmission, context-based resource selection, and fog computing extending toward aggregation and analytics.Context-based selection can combine resource accuracy, reliability, energy, cost, and availability; fog distributes compute, storage, control, and communication nearer users.Source: §6.2 · pp. 18–19

6.3 Data Aggregation and Integration: Interoperability

  • Interoperability enables aggregation and integration; surveyed approaches include semantic standards and ontologies, common data formats, and Linked Data and metadata models.The survey reports McKinsey’s estimate that interoperability could unlock 40 to 60 percent of projected future IoT market value.Source: §6.3 · p. 19

6.4 Architectures for Storage and Compute

  • Lambda-style batch, serving, and speed layers support analytics at scale, but Lambda, Hadoop, and Spark generally assume centralized or cloud-based clusters.The speed layer handles recent data and incremental updates to compensate for batch latency; the survey notes that these architectures suit decentralized fog and edge computing less well.Source: §6.4 · pp. 19–20

6.5 Storage Technologies

  • Distributed storage spans locally managed and remote-access file systems, federated databases and queries, message brokers, autonomous databases, and massively parallel processing databases.Federated query engines improve performance by optimizing join order, while data-distribution bottlenecks and other challenges remain; message brokers can provide reliable storage and delivery.Source: §6.5 · pp. 20–22

6.6 Compute Technologies and IoT Analytics Applications

  • Cloud compute options include virtual machines, serverless functions, and containers, while IoT applications use cloud analytics platforms and abstractions for distributed tasks.Container orchestration supports provisioning and managing distributed compute clusters; cloud time-series indexing can enable lower-latency real-time analytics at the cost of more storage.Source: §6.6 · p. 22
  • Distributed compute includes in-memory and stream systems, parallel and graph programming models, and edge/fog systems that schedule work near IoT devices.Spark uses micro-batches, whereas complex event processing uses continuous tuple-level operators; fog examples push stream-processing workloads to fog nodes.Source: §6.6 · pp. 22–23

6.7 Levels of Distribution of Storage and Compute

  • The survey classifies enabling technologies by storage and compute distribution across device, network, and cluster levels.The classification provides a reference for surveyed technologies; examples range from edge/fog systems to locally managed clusters.Source: §6.7 · p. 24, Figure 9 · p. 24
  • The review of data-flow infrastructure, storage and compute distribution, and distributed technologies provides a basis for future work addressing IoT big data analytics challenges.Source: §6.7 · p. 25

7 RESEARCH CHALLENGES

IoT analytics infrastructure faces tradeoffs among interoperability, performance, and analytical value. Variety complicates interoperability, while data volume and velocity complicate performance; scale remains an open challenge.

7 RESEARCH CHALLENGES

  • The survey frames infrastructure tradeoffs across distribution and interoperability, performance, and analytical value.Variety complicates interoperability; volume and velocity complicate performance, while analytical value concerns the output's position in the knowledge hierarchy.Source: §7 · p. 25
  • Semantic technologies can support interoperability, analytical value, and federation across heterogeneous sources, although system performance remains questionable.Edge and fog computing promise lower latency and efficiency through proximity to sources; cloud clusters and fast in-memory OLTP offer greater analytical value without IoT distribution and interoperability advantages.Source: §7 · p. 25
  • Integrating heterogeneous IoT and multimodal sources offers opportunities to enrich insights, improve predictive accuracy, and optimize prescription.The discussed sources include social media, Linked Data, images and video, satellite and geospatial data, and voice data.Source: §7 · p. 25
  • Performance and scalability questions extend beyond storage and communications to analytics processing, with distributed analytics proposed as a plausible response to IoT data scale.The paper predicts IoT data to be larger and richer than web-scale data and identifies distributed analytics as an area for further work.Source: §7 · pp. 25–26

8 CONCLUSIONS

The survey reviews IoT analytics applications and infrastructure, organizing analytics capabilities and application potential in a layered taxonomy. It identifies infrastructure tradeoffs as a basis for future research directions.

8 CONCLUSIONS

  • A systematic review of IoT analytics applications produces a layered taxonomy that categorizes analytics by capabilities and application potential.The taxonomy is intended to inform research and application roadmaps.Source: §8 · p. 26
  • The paper reviews enabling infrastructure and technologies across different stages of the analytics data flow.Source: §8 · p. 26
  • The survey discusses tradeoffs for IoT analytics as a basis for shaping future research directions.Source: §8 · p. 26
Loading 1807.00971v1…