Journal:DAQUA-MASS: An ISO 8000-61-based data quality management methodology for sensor data

Full article title	DAQUA-MASS: An ISO 8000-61-based data quality management methodology for sensor data
Journal	Sensors
Author(s)	Perez-Castillo, Ricardo; Carretero, Ana G.; Caballero, Ismael; Rodriguez, Moises; Piattini, Mario; Mate, Alejandro; Kim, Sunho; Lee, Dongwoo
Author affiliation(s)	University of Castilla-La Mancha, AQC Lab, University of Alicante, Myongji University, GTOne,
Primary contact	Email: ricardo dot pdelcastillo @ uclm dot es
Year published	2018
Volume and issue	18(9)
Page(s)	3105
DOI	10.3390/s18093105
ISSN	1424-8220
Distribution license	Creative Commons Attribution 4.0 International
Website	https://www.mdpi.com/1424-8220/18/9/3105/htm
Download	https://www.mdpi.com/1424-8220/18/9/3105/pdf (PDF)

Abstract

The internet of things (IoT) introduces several technical and managerial challenges when it comes to the use of data generated and exchanged by and between various smart, connected products (SCPs) that are part of an IoT system (i.e., physical, intelligent devices with sensors and actuators). Added to the volume and the heterogeneous exchange and consumption of data, it is paramount to assure that data quality levels are maintained in every step of the data chain/lifecycle. Otherwise, the system may fail to meet its expected function. While data quality (DQ) is a mature field, existing solutions are highly heterogeneous. Therefore, we propose that companies, developers, and vendors should align their data quality management mechanisms and artifacts with well-known best practices and standards, as for example, those provided by ISO 8000-61. This standard enables a process-approach to data quality management, overcoming the difficulties of isolated data quality activities. This paper introduces DAQUA-MASS, a methodology based on ISO 8000-61 for data quality management in sensor networks. The methodology consists of four steps according to the Plan-Do-Check-Act cycle by Deming.

Keywords: data quality; data quality management processes; ISO 8000-61; data quality in sensors; internet of things; IoT; smart, connected products; SCPs

Introduction

“Our economy, society, and survival aren’t based on ideas or information—they’re based on things.” This is one of the core foundations of the internet of things (IoT) as stated by Ashton, who coined the term. IoT is an emerging global internet-based information architecture facilitating the exchange of goods and services. IoT systems are inherently built on data gathered from heterogeneous sources in which the volume, variety, and velocity of data generation, exchanging and processing are dramatically increasing. Furthermore, there is a certain emergence of IoT semantic-oriented vision which needs ways to represent and manipulate the vast amount of raw data expected to be generated from and exchanged between the “things.”

The vast amount of data in IoT environments, gathered from a global-scale deployment of smart-things, is the basis for making intelligent decisions and providing better services (e.g., smart mobility, as presented by Zhang et al.). In other words, data represents the bridge that connects cyber and physical worlds. Despite of its tremendous relevance, if data are of inadequate quality, decisions from both humans and other devices are likely to be unsound. As a consequence, data quality (DQ) has become one of the key aspects in IoT. IoT devices, and in particular smart, connected products (SCPs), have concrete characteristics that favor the apparition of problems due to inadequate levels of data quality. Mühlhäuser defines SCPs as “entities (tangible object, software, or service) designed and made for self-organized embedding into different (smart) environments in the course of its lifecycle, providing improved simplicity and openness through improved connections.” While some of the SCP-related characteristics might be considered omnipresent (i.e., uncertain, erroneous, noisy, distributed, and voluminous), other characteristics are more specific and highly dependent on the context and monitored phenomena (i.e., smooth variation, continuous, correlation, periodicity, or Markovian behavior).

Also, outside of the IoT research area, DQ has been broadly studied during last years, and it has become a mature research area capturing the growing interest of the industry due to the different types of values that companies can extract from data. This fact is reflected by the standardization efforts like ISO/IEC 25000 series addressing systems and software quality requirements and evaluation (SQuaRE) processes, and specific techniques for managing data concerns. We pose that such standards can be tailored and used within the IoT context, not only bring benefits standardizing solutions and enabling a better communication between partners. Also, the number of problems and system fails on the IoT environment is reduced, better decisions can be taken due to a better quality of data, all stakeholders are aligned and can take benefit of the advances on the standard used, and it is easier to apply data quality solutions in a global way because the heterogeneity is reduced.

Due to the youth of IoT, and despite DQ standards, frameworks, management techniques, and tools proposed in the literature, DQ for IoT has not been yet widely studied. However, and prior to this research line, it is possible to cite some works that had addressed some DQ concerns in sensor wireless networks, or in data streaming among other proposals. However, these works have not considered the management of DQ in a holistic way in line with existing DQ-related standards. In our attempt to align the study of DQ in IoT to international standards, this paper provides practitioners and researchers with DAQUA-MASS, a methodology for managing data quality in SCP environments, which considers some of the DQ best practices for improving quality of data in SCP environments aligned to ISO 8000-61. Due to the intrinsic distributed nature of IoT systems, using such standards will enable the various organizations to be aligned to the same foundations, and in the end, to work in a seamless way, what will undoubtedly improve the performance of the business processes.

The remainder of this paper is organized as follows: the next section presents the most challenging data quality management concerns in the context of the SCP environments; afterwards. related work is explored. Then the data quality model in which our methodology is based on is presented. The last two sections propose a methodology for managing data quality in SCP environments and discuss conclusions and implications of this work.

Data quality challenges in SCP environments

This section introduces some general ideas about SCPs and their operation as an essential part of IoT. In addition, some challenges related to DQ in SCP environments are also introduced.

According to Cook and Das, a smart environment is a small world where all kinds of smart devices are continuously working to make inhabitants’ lives more comfortable. According to Mühlhäuser, SCP provides intelligent actions through improved connections by means of context-awareness, semantic self-description, proactive behavior, multimodal natural interfaces, AI planning, and machine learning.

SCPs have three main core components: physical, smart, and connectivity components. Smart components extend the capabilities and value of the physical components, while connectivity extends the capabilities and value of the smart components. This enables some smart components to exist outside the physical product itself, with a cycle of value improvement.

IoT and SCP can be confused in some contexts. However, IoT simply reflects the growing number of SCPs and highlights the new opportunities they can represent. IoT, which can involve people or things, is a means for interchanging information. What makes SCPs essentially different is not the "internet," but the changing nature of the “things.” A product that is smart and connected to the cloud could become part of an interconnected management solution, and companies can therefore evolve from making products to offering more complex, higher-value services within a “system of systems.”

SCPs include processors, sensors, software, and connectivity that allow data to be exchanged between the product and its environment. The data collected by sensors of these SCPs can be then analysed to inform decision-making, enable operational efficiencies, and continuously improve the performance of the product. This paper focuses on the data produced by such sensors, and how inadequate levels of data quality may affect the processing of the data, while smart and connectivity parts of SCPs are outside of the scope of this paper.

SCPs can be connected in large, complex networks throughout three different layers: acquisition, processing, and utilization layers (see Figure 1).

Figure 1. Layers in SCP environments.

• The acquisition layer refers to the sensor data collection system where sensors, raw (or sensed) data, and pre-processed data are managed. This is the main focus of this paper.
• The processing layer involves data resulting from the data processing and management center, where energy, storage, and analysis capabilities are more significant.
• The utilization layer concerns delivered data (or post-processed data) exploited, for example, over a geographic information system (GIS) or combined with other services or applications.

As previously stated, the scope of the paper is limited to the data produced by SCPs’ sensors. Hence, the proposal is mainly intended to be applied in the context of the acquisition layer. Nevertheless, the management of data quality in sensors can impact on how data is processed (processing layer) and how data may be used later (utilization layer).

Networking and management of SCP operations can generate the business intelligence needed to deliver smart services. Smart services are delivered to or via smart objects that feature awareness and connectivity. SCP can carry out the following functions to support smart services: status, diagnostics, upgrades, control and automation, profiling and behavior tracking, replenishment and commerce, location mapping. and logistics, among others.

SCP operations enable new capabilities for companies, although new problems and challenges that arise must also be taken into account. On one hand, SCP operations require companies to build and support an entirely new technology infrastructure. Technological layers in the new technology landscape include new product hardware, embedded software, connectivity, a product cloud running on remote servers, security tools, gateway for external information sources, and integration with enterprise business systems. On the other hand, SCP operations can provide competitive advantages, which are based on the operational effectiveness. Operation effectiveness requires the embrace of best practices along the value chain, including up-to-date product technologies, the latest production equipment, and state-of-the-art sales force methods, IT solutions, and so forth. Thus, SCP operations also creates new best practices across the value chain.

According to the different sources of data in these SCP environments, we can distinguish different types of aggregated data:

• Sensor data: data that is generated by sensors and digitized in a computer-readable format (for example, camera sensor readings)
• Device data: integrated by sensor data; observed metadata (metadata that characterizes the sensor data, e.g., timestamp of sensor data); and device meta data (metadata that characterizes the device, e.g., device model, sensor model, manufacturer, etc.), so device data, for example, can be data coming from the camera (device)
• General data: data related to/or coming from devices which has been modified or computed to derive different data plus business data (i.e., data for business use such as operation, maintenance, service, customers, etc.)
• IoT data: general data plus device data

A reduction in the levels of quality of these data due to different problems in SCP operations can threaten the success factors of SCP environments. The quality of produced data is often affected by dysfunctional SCP devices and sensors, which are the sources providing data, and can potentially result in inadequate levels of quality that are only detected later on, when data are being processed and used. Therefore, while we can identify dysfunctional SCP devices through the analysis of sensor data by using data quality management techniques, it is noteworthy that these devices will impact the rest of the sensor network. According to, Table 1 summarizes some of these SCP factors that, in some cases, could condition or lead to data quality issues. In addition, the three columns on the right of Table 1 show (marked with a cross) the most critical layers affected in a greater extent by every SCP factor.

Table 1. SCP factors that can finally affect the levels of DQ according to Karkouch et al. SCP Factor Side Effect in Data Quality Acquisition Processing Utilization Deployment scale SCPs are expected to be deployed on a global scale. This leads to a huge heterogeneity in data sources (not only computers but also daily objects). Also, the huge number of devices accumulates the chance of error occurrence. X X Resource constraints For example, computational and storage capabilities that do not allow complex operations due, in turn, to the battery-power constraints among others. X X Network Intermittent loss of connection in the IoT is recurrent. Things are only capable of transmitting small-sized messages due to their scarce resources. X X Sensors Embedded sensors may lack precision or suffer from loss of calibration or even low accuracy. Faulty sensors may also result in inconsistencies in data sensing. X Environment SCP devices will not be deployed only in tolerant and less aggressive environments. To monitor some phenomenon, sensors may be deployed in environments with extreme conditions. Data errors emerge when the sensor experiences the surrounding environment influences. X X Vandalism Things are generally defenseless from outside physical threats (both from humans and animals). X X Fail-dirty A sensor node fails, but it keeps up reporting readings which are erroneous. It is a common problem for SCP networks and an important source of outlier readings. X X Privacy Privacy preservation processing, thus DQ could be intentionally reduced. X Security vulnerability Sensor devices are vulnerable to attack, e.g., it is possible for a malicious entity to alter data in an SCP device. X X Data stream processing Data gathered by smart things are sent in the form of streams to the back-end pervasive applications which make use of them. Some stream processing operators could affect quality of the underlying data. Other important factors are data granularity and variety. Granularity concerns interpolation and spatio-temporal density while variety refers to interoperability and dynamic semantics. X X

Tilak et al. provide a taxonomy of sensor errors. These errors are directly related to different data quality problems in the acquisition layer. The mentioned taxonomy distinguishes the following six types of data sensors errors (see Table 2). Apart from errors in isolated SCP devices, there are other communication errors which can happen at the SCP network level. Table 3 summarizes the main types of communication errors: omission, crashes, delay, and message corruption. The table shows the DQ issue derived by each problem, the root cause, and possible solution.

Table 2. Sensors errors deriving DQ problems in SCP environments (adapted from Jesus et al.)

Table 3. SCP Network Errors Sensor fault DQ problem Root cause Solution Omission faults Absence of values Missing sensor Network reliability, retransmission Crash faults (fading/intermittent) Inaccuracy/absence of values Environment interference Redundancy/estimating with past values Delay faults Inaccuracy Time domain Timeline solutions Message corruption Integrity Communication Integrity validation

All the mentioned SCP devices/sensor errors will lead to different DQ problems in the three layers depicted in Figure 1. As previously mentioned, DQ problems can be represented as a degradation of some DQ characteristics that are especially important in different environments. Let us consider two groups of data quality characteristics:

• DQ characteristics to assess data quality in use within a specific context. This aspect considers selected criteria to estimate the quality of raw sensor data at the acquisition and processing layer. There are some DQ characteristics considered, which make it possible to estimate the quality on data sources, their context of acquisition, and their transmission to the data management and processing. These DQ characteristics are accuracy and completeness according to ISO/IEC 25012 and reliability and communication reliability as proposed by Rodriguez and Servigne. It is also related to the utilization layer and includes availability regarding ISO/IEC 25012 plus timeliness and adequacy as defined by Rodriguez and Servigne.
• DQ Characteristics aimed at managing internal data quality. The main goal of managing internal data quality is to avoid inconsistent data and maintain the temporality of sensor data at the processing layer. These characteristics are consistency and currency according to ISO/IEC 25012 and volatility as proposed by Rodriguez and Servigne.

Related work

The goal of this section is twofold. First, works related to the study of data quality in sensor networks and SCP environments in general are presented. Second, data quality methodologies are introduced and compared in order to draw the main contribution of the proposed methodology.

Sensor data quality

There are some published works in the literature that address the concerns related to data quality management in SCP and IoT environments. Karkouch et al. presented a state-of-the-art survey for DQ in IoT. This survey presents IoT-related factors endangering the DQ and their impact on various DQ characteristics. Also, DQ problem manifestations are discussed (and their symptoms identified) as well as their impact in the context of IoT. Jesus et al. provided a similar survey addressing the problem of not being able to ensure desired DQ levels for dependable monitoring when using wireless sensor networks. This work pays special attention to comprehension of which faults can affect sensors, how they can affect the quality of the information, and how this quality can be improved and quantified. Rodriguez and Servigne also analysed data errors in sensor networks, in particular in environmental monitoring systems. In this paper, authors address the problem of uncertainty of data coming from sensors with an approach dedicated to providing environmental monitoring applications and users with data quality information. Badawy et al. combined parametric and non-parametric signal processing and machine learning algorithms for automating sensor data quality control, which can identify those parts of the sensor data that are sufficiently reliable for further analysis and discards useless data.

Another research subarea of DQ in sensor networks is the DQ management in sensor data streams. Klein et al. presented a quality-driven load shedding approach that screens the data stream to find and discard data items of minor quality. Thus, DQ of stream processing results is maximized under adverse conditions such as data overload. Campbell et al. advocated for automated quality assurance and quality control procedures based on graphical and statistical summaries for review and track the provenance of the data in environmental sensor streams.

Other works focus on data management from different viewpoints. For example, Al-Ruithe et al. detailed roles, responsibilities, and policies in the context of IoT-cloud converged environments and provide a generic framework for data governance and security. Similarly, Qin et al. provided a data management perspective on large-scale sensor environment applications posing non-functional requirements to meet the underlying timeliness, reliability, and accuracy needs in addition to the functional needs of data collection. Although all these approaches are interesting and provide a useful vision, they still do not address how to make available (e.g., institutionalize) best practices in data quality management to the entire organization. Such an approach has been proven to be more efficient when it comes to creating an organizational data quality culture. This vision is specifically important in the case of IoT, since the SCP operations can be executed across different networks belonging to different departments or even organizations. From our point of view, this is a critical aspect for IoT that must be covered in a holistic way.

Data quality methodologies comparison

There are some methodologies that can be used as drivers for assessing and managing DQ. First, Lee et al. proposed AIMQ as a methodology that encompasses a model of data quality, a questionnaire to measure DQ, and analysis techniques for interpreting the DQ measures. This methodology is mainly used to analyse the gap between an organization and best practices, as well as to assess gaps between information systems professionals and data consumers. The application of this methodology is useful for determining the best area for DQ improvement activities. This methodology has not been widely used and in a greater extent it has been considered to be too theoretical and dependent on the domain. McGilvray provides a practical approach for planning and managing information quality. In comparison with the methodology proposed by Lee et al., McGilvray provides a more pragmatic and practical approach to achieving the desired state of DQ within an organization. However, this methodology is still dependent on the domain of application. ISO/TS 8000-150:2011 “specifies fundamental principles of master data quality management, and requirements for implementation, data exchange and provenance.” This standard constitutes an informative framework that identifies processes for DQ management. This framework could be used in conjunction with, or independently of, quality management systems standards, for example, ISO 9001.

Batini et al. provided a literature review about different methodologies for data quality assessment and improvement. Most of the methods and techniques included in such review cannot be considered as DQ management methodologies since they do not consider all the managerial concerns in a holistic manner. On the contrary, most of these methods are focused on DQ assessment or improvement in isolation. Similar to the mentioned review, a more recent study developed by Woodall et al. classified most recent DQ assessment and improvement methods. This work suffers the same problem that the work of Batini et al. does. Apart from these examples, there is a lack of comprehensive methodologies for the assessment and improvement of DQ in the domain of SCP operations and their underlaying data.

The recent standard ISO 8000-61 provides a set of standard guidelines for managing DQ in a holistic way, which can be tailored for different domains. However, its main purpose is not to serve as a methodology for DQ management per se, but it simply provides a process reference model. In this sense, the standard is more descriptive than operative, what makes it not usable out-of-the-box. Aligned with this standard, this paper proposes the DAQUA-MASS methodology to deal with DQ in SCP environments. The main contribution of the DAQUA-MASS methodology is that it takes the standard best practices for depicting an operative way to manage DQ (as depicted in the processes of ISO 8000-61) and tailors these to the particular domain of SCP environments, and in particular, in sensor-related data.

DAQUA-Model: A data quality model

Reviewing the literature, it is possible to find that the concept of "data quality" has been defined in different ways. The widest used definitions are aligned with the concept of “fitness for use.” Strong et al. define data quality as "data that is fit for use by data consumer. This means that usefulness and usability are important aspects of quality”. Different stakeholders can have different perceptions of what quality means for the same data. It largely depends on the context in which data is used. Thus, DQ in IoT environments must be adequately managed considering the very nature of the IoT systems. Typically, to improve data quality, a Plan-Do-Check-Act (PDCA) cycle specifically tailored for the context of usage is followed. In this sense, we think that adopting the processes of ISO 8000-61 which are deployed in the PDCA order can largely help to provide a methodology for managing data quality in IoT environments. At the core of the PDCA cycle for IoT environments is the identification of a Data Quality Model (DQModel) which, being composed of several data quality characteristics suitable for the problem, is used to identify and represent the data quality requirements required in the context.

In our case, and according to our philosophy of aligning with international standards, the DQ Model proposed is a specialization of the DQ Model introduced in ISO/IEC 25012. This DQ Model is widely accepted and used in the industry. Nevertheless, it has not been specifically developed for considering SCP aspects. In fact, the scope section of such standard literally states that it “does not include data produced by embedded devices or real time sensors that are not retained for further processing or historical purposes.” Therefore, in order to complement the standard, we provide some orientation in this paper on how to specifically use ISO/IEC 25012 in the context of SCP environments.

The DQ Model focuses on the quality of the data as part of an information system and defines quality characteristics for target data used by humans and systems (i.e., the data that the organization decides to analyse and validate through the model). This model categorizes quality attributes into 15 characteristics and considers three perspectives: inherent, system-dependent, and both jointly (see crosses in Table 4).

Table 4. DQ Characteristics in ISO/IEC 25012 that can be affected by sensor data errors

Notes

This presentation is faithful to the original, with only a few minor changes to presentation. Grammar was cleaned up for smoother reading. In some cases important information was missing from the references, and that information was added.