Attacking Quality Issues in Data Warehousing

[magazine]
Volume-Issue: 
Summary:

To fully detect, isolate, and resolve quality issues in a traditional, large-scale data warehouse requires that several approaches be used together. Wayne identifies types of data quality issues and then illustrates how to best attack and resolve those pesky issues.

Implementation and growth of data warehouses continue to gain attraction as organizations become more aware of the benefits of decision and analytic-oriented databases. Nevertheless, there is often one important obstacle to the rapid development of commercial data warehouses: data quality. Serious problems are often discovered when planning and populating a warehouse that, if not resolved, can delay or eventually result in terminating the project.

During the past twenty years, researchers have contributed to the understanding of data quality issues, yet little research has been collectively compiled to identify root causes of data quality problems that occur throughout major phases of data warehousing.

Based on my experience, the following are primary causes of data quality defects in data warehousing [1]:

• Flaws in the data warehouse modeling and schema design
• Defects in data sources used as input to the data warehouse
• Failure to effectively profile source and target data
• Weaknesses in the design and implementation process for data warehouse staging and extract, transform, and load processes

Using Early-Phase Defect Prevention Methods

This article highlights the reasons for data deficiencies related to the root causes listed above together with timely quality assurance efforts that can be implemented for discovery and correction. It is hoped that data warehouse designers, developers, and testers cooperate and benefit by examining these quality issues before moving forward with data integration into the data warehouse.

Figure 1 displays a high-level view of the common data warehousing extract, transform, and load process where data quality and functional testing are recommended.

Figure 1

Data Quality Issues Related to Data Model and Schema Design

Design of the data model for the data warehouse greatly influences the quality of the analysis by programs that use the data. A flawed schema will negatively impact information quality.

Data modeling is the process used to define and analyze data requirements needed to support business processes within the scope of application needs. The data modeling process should involve trained data modelers working closely with business stakeholders, developers, quality assurance, and potential users of the information system. Data modeling defines not only data elements, but also their  structures and the relationships between them.

Data modeling methodologies should be used to model data in a standard, consistent, and predictable manner in order to manage the data as a resource. The use of modern data modeling standards and tools is strongly recommended for all projects.

Table 1 shows highlights of how a proper data warehouse design review can make or break your data warehouse. [2]

Quality Issues in the Data Warehouse Source Data

A leading cause of data warehousing and business intelligence project failures is finding and then loading incorrect or poor-quality data. The source system often consists of transaction and production raw  data, which is where the details are pulled from and made suitable for the data warehouse. Each of these data sources usually has its own diverse methods of storing data, which may contribute to data  quality problems if proper care is not taken.

Data warehouse environments provide the source of information used by business units to make strategic decisions. However, much of that data is created outside the warehouse. That means data quality  problems can originate at the source and can therefore persist due to faulty data acquisition and delivery processes, or interpretation and transformation glitches.

Data quality problems in source systems need to be recognized as requiring mitigation. This can be accomplished by either addressing these problems as defects or by getting approval by stakeholders  that these issues are acceptable. The QA team must then ensure that data warehouse users are aware of these data quality deficiencies in cases where they are not fixed before being loaded into the data warehouse.

Under certain conditions, source files are the product of multiple file consolidations. Consolidated files can, in turn, result in data quality being compromised before being loaded into the data warehouse staging area. Table 2 summarizes a few other possible causes of data quality issues as data sources are staged into the warehouse.

Other reasons for data pollution issues in the data warehouse may be cases where data was never being fully captured by source systems, the use of heterogeneous system integrations, and a failure to have an adequate policy for data warehouse project planning.

Table 1

Table 2

Discovering Data Quality Issues Using Data Profiling Techniques

When potential data sources are identified and then finalized and agreed to, data profiling should be implemented immediately on that source data. Data profiling is the examination and assessment of your source systems’ data quality, integrity, and consistency—sometimes known as source systems analysis. [3]

As important as data profiling is, it is often ignored and, as a result, data warehouse quality can be significantly compromised.

The data quality assurance analyst supports an organization’s data quality initiatives by analyzing data. Profiling is the primary method for performing a data quality assessment. Data profiling is also  used to quantify the extent of problems surfaced by other means and to measure the impact that data quality remedies have had.

Listed below are examples of problems that are easily uncovered through data profiling:

• Data fields used for purposes other than expected
• Fields that contain no data for any record
• Missing values when a field is defined as NOT NULL
• Violations of business rules

Business analysts can determine problem root causes during data analysis that could result in a substantial number of data quality problems that need to be corrected.

At the beginning of a data warehouse project and as soon as potential data sources are identified, data profiling assessments should be conducted to prepare for a go/no-go decision about proceeding with the project.

Table 3 depicts just a few of the possible causes of data quality degradation discovered at the profiling stage of data warehousing.

Table 3

 

Table 4

Data Quality Issues Discovered During Data Loading

An important design consideration is whether data cleansing should be conducted for each source input during the staging phase, during the extract, transform, and load process, or within the data  warehouse. The data staging area is where “grooming” is often conducted on data after it is loaded from source systems. [4]

Data staging and the extract, transform, and load phases are considered to be the most crucial phases of data warehousing, where maximum responsibility for data quality efforts occurs. These are prime  phases for validating data quality from sources or auditing and discovering data issues. There may be several reasons for data quality problems during the staging and extract, transform, and load  phases. A few of those are listed in table 4.

When data quality problems are encountered while importing data into the data warehouse, there are four viable actions that can be taken: exclude the data, accept the data, correct the data, or insert a  default value. These are some of the design decisions that must be faced while working to improve data quality in early phases of data warehouse projects.

In Summary

There are many causes of data quality problems that may be found throughout all phases of data warehouse development. Data quality issues have been classified and described in a way that should help data warehouse practitioners, implementers, and tool providers find and resolve these issues as they move forward with each phase of data warehousing. {end} [email protected]

References

Austin, Matt. “The Necessity of Data Profiling: A How-to Guide to Getting Started and Driving Value.” TDWI. February 3, 2010. http://tdwi.org/articles/2010/02/03/data-profiling-value.aspx

Geiger, Jonathan G. “Data Quality Management, The Most Critical Initiative You Can Implement.” SUGI 29. May 2004: Paper 098-29. http://www2.sas.com/proceedings/sugi29/098-29.pdf.

Inmon, W.H. Building the Data Warehouse . 4th ed. Indianapolis: Wiley Computer Publishing, 2005.

Singh, Ranjit and Dr. Kawaljeet Singh, “A Descriptive Classification of Causes of Data Quality Problems in Data Warehousing.” International Journal of Computer Science Issues , vol. 7, issue 3, no. 2, May 2010. http://ijcsi.org/papers/7-3-2-41-50.pdf

About the author

Upcoming Events

Apr 28
Jun 02
Sep 22
Oct 13