An Introduction to Data Mining

Discovering hidden value in your data warehouse

By Kurt Thearling

Overview

Data mining, the extraction of hidden predictive information from large databases, is
a powerful new technology with great potential to help companies focus on the most
important information in their data warehouses. Data mining tools predict future
trends and behaviors, allowing businesses to make proactive, knowledge-driven
decisions. The automated, prospective analyses offered by data mining move beyond
the analyses of past events provided by retrospective tools typical of decision support
systems. Data mining tools can answer business questions that traditionally were too
time consuming to resolve. They scour databases for hidden patterns, finding
predictive information that experts may miss because it lies outside their expectations.

Most companies already collect and refine massive quantities of data. Data mining
techniques can be implemented rapidly on existing software and hardware platforms
to enhance the value of existing information resources, and can be integrated with new
products and systems as they are brought on-line. When implemented on high
performance client/server or parallel processing computers, data mining tools can
analyze massive databases to deliver answers to questions such as, "Which clients are
most likely to respond to my next promotional mailing, and why?"

This white paper provides an introduction to the basic technologies of data mining.
Examples of profitable applications illustrate its relevance to today’s business
environment as well as a basic description of how data warehouse architectures can
evolve to deliver the value of data mining to end users.

The Foundations of Data Mining

Data mining techniques are the result of a long process of research and product
development. This evolution began when business data was first stored on computers,
continued with improvements in data access, and more recently, generated
technologies that allow users to navigate through their data in real time. Data mining
takes this evolutionary process beyond retrospective data access and navigation to
prospective and proactive information delivery. Data mining is ready for application
in the business community because it is supported by three technologies that are now
sufficiently mature:
  Massive data collection
 Powerful multiprocessor computers
 Data mining algorithms

Commercial databases are growing at unprecedented rates. A recent META Group

survey of data warehouse projects found that 19% of respondents are beyond the 50
gigabyte level, while 59% expect to be there by second quarter of 1996.1 In some
industries, such as retail, these numbers can be much larger. The accompanying need
for improved computational engines can now be met in a cost-effective manner with
parallel multiprocessor computer technology. Data mining algorithms embody
techniques that have existed for at least 10 years, but have only recently been
implemented as mature, reliable, understandable tools that consistently outperform
older statistical methods.

In the evolution from business data to business information, each new step has built
upon the previous one. For example, dynamic data access is critical for drill-through
in data navigation applications, and the ability to store large databases is critical to
data mining. From the user’s point of view, the four steps listed in Table 1 were
revolutionary because they allowed new business questions to be answered accurately
and quickly.

Evolutionary Business Question Enabling Product Characteristics

Step Technologies Providers

Data Collection "What was my total Computers, tapes, IBM, CDC Retrospective,
revenue in the last five disks static data
(1960s) years?" delivery

Data Access "What were unit sales Relational databases Oracle, Retrospective,
in New England last (RDBMS), Sybase, dynamic data
(1980s) March?" Structured Query Informix, delivery at
Language (SQL), IBM, record level
ODBC Microsoft

Data "What were unit sales On-line analytic Pilot, Retrospective,

Warehousing & in New England last processing (OLAP), Comshare, dynamic data
March? Drill down to multidimensional Arbor, delivery at
Decision Boston." databases, data Cognos, multiple levels
Support warehouses Microstrategy

(1990s)
Data Mining "What’s likely to Advanced Pilot, Prospective,
happen to Boston unit algorithms, Lockheed, proactive
(Emerging sales next month? multiprocessor IBM, SGI, information
Today) Why?" computers, massive numerous delivery
databases startups
(nascent
industry)

Table 1. Steps in the Evolution of Data Mining.

The core components of data mining technology have been under development for
decades, in research areas such as statistics, artificial intelligence, and machine
learning. Today, the maturity of these techniques, coupled with high-performance
relational database engines and broad data integration efforts, make these technologies
practical for current data warehouse environments.

The Scope of Data Mining

Data mining derives its name from the similarities between searching for valuable
business information in a large database — for example, finding linked products in
gigabytes of store scanner data — and mining a mountain for a vein of valuable ore.
Both processes require either sifting through an immense amount of material, or
intelligently probing it to find exactly where the value resides. Given databases of
sufficient size and quality, data mining technology can generate new business
opportunities by providing these capabilities:

 Automated prediction of trends and behaviors. Data mining automates the

process of finding predictive information in large databases. Questions that
traditionally required extensive hands-on analysis can now be answered
directly from the data — quickly. A typical example of a predictive problem is
targeted marketing. Data mining uses data on past promotional mailings to
identify the targets most likely to maximize return on investment in future
mailings. Other predictive problems include forecasting bankruptcy and other
forms of default, and identifying segments of a population likely to respond
similarly to given events.

 Automated discovery of previously unknown patterns. Data mining tools

sweep through databases and identify previously hidden patterns in one step.
An example of pattern discovery is the analysis of retail sales data to identify
seemingly unrelated products that are often purchased together. Other pattern
discovery problems include detecting fraudulent credit card transactions and
identifying anomalous data that could represent data entry keying errors.
Data mining techniques can yield the benefits of automation on existing software and
hardware platforms, and can be implemented on new systems as existing platforms
are upgraded and new products developed. When data mining tools are implemented
on high performance parallel processing systems, they can analyze massive databases
in minutes. Faster processing means that users can automatically experiment with
more models to understand complex data. High speed makes it practical for users to
analyze huge quantities of data. Larger databases, in turn, yield improved predictions.

Databases can be larger in both depth and breadth:

 More columns. Analysts must often limit the number of variables they
examine when doing hands-on analysis due to time constraints. Yet variables
that are discarded because they seem unimportant may carry information about
unknown patterns. High performance data mining allows users to explore the
full depth of a database, without preselecting a subset of variables.

 More rows. Larger samples yield lower estimation errors and variance, and
allow users to make inferences about small but important segments of a
population.

A recent Gartner Group Advanced Technology Research Note listed data mining and
artificial intelligence at the top of the five key technology areas that "will clearly have
a major impact across a wide range of industries within the next 3 to 5 years."2
Gartner also listed parallel architectures and data mining as two of the top 10 new
technologies in which companies will invest during the next 5 years. According to a
recent Gartner HPC Research Note, "With the rapid advance in data capture,
transmission and storage, large-systems users will increasingly need to implement
new and innovative ways to mine the after-market value of their vast stores of detail
data, employing MPP [massively parallel processing] systems to create new sources
of business advantage (0.9 probability)."3

The most commonly used techniques in data mining are:

 Artificial neural networks: Non-linear predictive models that learn through

training and resemble biological neural networks in structure.

 Decision trees: Tree-shaped structures that represent sets of decisions. These

decisions generate rules for the classification of a dataset. Specific decision tree
methods include Classification and Regression Trees (CART) and Chi Square
Automatic Interaction Detection (CHAID) .
  Genetic algorithms: Optimization techniques that use processes such as
genetic combination, mutation, and natural selection in a design based on the
concepts of evolution.

 Nearest neighbor method: A technique that classifies each record in a dataset

based on a combination of the classes of the k record(s) most similar to it in a
historical dataset (where k ³ 1). Sometimes called the k-nearest neighbor
technique.

 Rule induction: The extraction of useful if-then rules from data based on
statistical significance.

Many of these technologies have been in use for more than a decade in specialized
analysis tools that work with relatively small volumes of data. These capabilities are
now evolving to integrate directly with industry-standard data warehouse and OLAP
platforms. The appendix to this white paper provides a glossary of data mining terms.

How Data Mining Works

How exactly is data mining able to tell you important things that you didn't know or
what is going to happen next? The technique that is used to perform these feats in data
mining is called modeling. Modeling is simply the act of building a model in one
situation where you know the answer and then applying it to another situation that you
don't. For instance, if you were looking for a sunken Spanish galleon on the high seas
the first thing you might do is to research the times when Spanish treasure had been
found by others in the past. You might note that these ships often tend to be found off
the coast of Bermuda and that there are certain characteristics to the ocean currents,
and certain routes that have likely been taken by the ship’s captains in that era. You
note these similarities and build a model that includes the characteristics that are
common to the locations of these sunken treasures. With these models in hand you
sail off looking for treasure where your model indicates it most likely might be given
a similar situation in the past. Hopefully, if you've got a good model, you find your
treasure.

This act of model building is thus something that people have been doing for a long
time, certainly before the advent of computers or data mining technology. What
happens on computers, however, is not much different than the way people build
models. Computers are loaded up with lots of information about a variety of situations
where an answer is known and then the data mining software on the computer must
run through that data and distill the characteristics of the data that should go into the
model. Once the model is built it can then be used in similar situations where you
don't know the answer. For example, say that you are the director of marketing for a
telecommunications company and you'd like to acquire some new long distance phone
customers. You could just randomly go out and mail coupons to the general
population - just as you could randomly sail the seas looking for sunken treasure. In
neither case would you achieve the results you desired and of course you have the
opportunity to do much better than random - you could use your business experience
stored in your database to build a model.

As the marketing director you have access to a lot of information about all of your
customers: their age, sex, credit history and long distance calling usage. The good
news is that you also have a lot of information about your prospective customers: their
age, sex, credit history etc. Your problem is that you don't know the long distance
calling usage of these prospects (since they are most likely now customers of your
competition). You'd like to concentrate on those prospects who have large amounts of
long distance usage. You can accomplish this by building a model. Table 2 illustrates
the data used for building a model for new customer prospecting in a data warehouse.

Customers Prospects

General information (e.g. demographic Known Known

data)

Proprietary information (e.g. customer Known Target

transactions)

Table 2 - Data Mining for Prospecting

The goal in prospecting is to make some calculated guesses about the information in
the lower right hand quadrant based on the model that we build going from Customer
General Information to Customer Proprietary Information. For instance, a simple
model for a telecommunications company might be:

98% of my customers who make more than $60,000/year spend more than $80/month
on long distance

This model could then be applied to the prospect data to try to tell something about
the proprietary information that this telecommunications company does not currently
have access to. With this model in hand new customers can be selectively targeted.
Test marketing is an excellent source of data for this kind of modeling. Mining the
results of a test market representing a broad but relatively small sample of prospects
can provide a foundation for identifying good prospects in the overall market. Table 3
shows another common scenario for building models: predict what is going to happen
in the future.

Yesterday Today Tomorrow

Static information and current Known Known Known

plans (e.g. demographic data,
marketing plans)

Dynamic information (e.g. Known Known Target

customer transactions)

Table 3 - Data Mining for Predictions

If someone told you that he had a model that could predict customer usage how would
you know if he really had a good model? The first thing you might try would be to ask
him to apply his model to your customer base - where you already knew the answer.
With data mining, the best way to accomplish this is by setting aside some of your
data in a vault to isolate it from the mining process. Once the mining is complete, the
results can be tested against the data held in the vault to confirm the model’s validity.
If the model works, its observations should hold for the vaulted data.

An Architecture for Data Mining

To best apply these advanced techniques, they must be fully integrated with a data
warehouse as well as flexible interactive business analysis tools. Many data mining
tools currently operate outside of the warehouse, requiring extra steps for extracting,
importing, and analyzing the data. Furthermore, when new insights require
operational implementation, integration with the warehouse simplifies the application
of results from data mining. The resulting analytic data warehouse can be applied to
improve business processes throughout the organization, in areas such as promotional
campaign management, fraud detection, new product rollout, and so on. Figure 1
illustrates an architecture for advanced analysis in a large data warehouse.
 Figure 1 - Integrated Data Mining Architecture

The ideal starting point is a data warehouse containing a combination of internal data
tracking all customer contact coupled with external market data about competitor
activity. Background information on potential customers also provides an excellent
basis for prospecting. This warehouse can be implemented in a variety of relational
database systems: Sybase, Oracle, Redbrick, and so on, and should be optimized for
flexible and fast data access.

An OLAP (On-Line Analytical Processing) server enables a more sophisticated end-

user business model to be applied when navigating the data warehouse. The
multidimensional structures allow the user to analyze the data as they want to view
their business – summarizing by product line, region, and other key perspectives of
their business. The Data Mining Server must be integrated with the data warehouse
and the OLAP server to embed ROI-focused business analysis directly into this
infrastructure. An advanced, process-centric metadata template defines the data
mining objectives for specific business issues like campaign management,
prospecting, and promotion optimization. Integration with the data warehouse enables
operational decisions to be directly implemented and tracked. As the warehouse grows
with new decisions and results, the organization can continually mine the best
practices and apply them to future decisions.

This design represents a fundamental shift from conventional decision support

systems. Rather than simply delivering data to the end user through query and
reporting software, the Advanced Analysis Server applies users’ business models
directly to the warehouse and returns a proactive analysis of the most relevant
information. These results enhance the metadata in the OLAP Server by providing a
dynamic metadata layer that represents a distilled view of the data. Reporting,
visualization, and other analysis tools can then be applied to plan future actions and
confirm the impact of those plans.
Profitable Applications

A wide range of companies have deployed successful applications of data mining.

While early adopters of this technology have tended to be in information-intensive
industries such as financial services and direct mail marketing, the technology is
applicable to any company looking to leverage a large data warehouse to better
manage their customer relationships. Two critical factors for success with data mining
are: a large, well-integrated data warehouse and a well-defined understanding of the
business process within which data mining is to be applied (such as customer
prospecting, retention, campaign management, and so on).

Some successful application areas include:

 A pharmaceutical company can analyze its recent sales force activity and their
results to improve targeting of high-value physicians and determine which
marketing activities will have the greatest impact in the next few months. The
data needs to include competitor market activity as well as information about
the local health care systems. The results can be distributed to the sales force
via a wide-area network that enables the representatives to review the
recommendations from the perspective of the key attributes in the decision
process. The ongoing, dynamic analysis of the data warehouse allows best
practices from throughout the organization to be applied in specific sales
situations.
 A credit card company can leverage its vast warehouse of customer transaction
data to identify customers most likely to be interested in a new credit product.
Using a small test mailing, the attributes of customers with an affinity for the
product can be identified. Recent projects have indicated more than a 20-fold
decrease in costs for targeted mailing campaigns over conventional approaches.
 A diversified transportation company with a large direct sales force can apply
data mining to identify the best prospects for its services. Using data mining to
analyze its own customer experience, this company can build a unique
segmentation identifying the attributes of high-value prospects. Applying this
segmentation to a general business database such as those provided by Dun &
Bradstreet can yield a prioritized list of prospects by region.
 A large consumer package goods company can apply data mining to improve
its sales process to retailers. Data from consumer panels, shipments, and
competitor activity can be applied to understand the reasons for brand and store
switching. Through this analysis, the manufacturer can select promotional
strategies that best reach their target customer segments.

Each of these examples have a clear common ground. They leverage the knowledge
about customers implicit in a data warehouse to reduce costs and improve the value of
customer relationships. These organizations can now focus their efforts on the most
important (profitable) customers and prospects, and design targeted marketing
strategies to best reach them.

Conclusion

Comprehensive data warehouses that integrate operational data with customer,

supplier, and market information have resulted in an explosion of information.
Competition requires timely and sophisticated analysis on an integrated view of the
data. However, there is a growing gap between more powerful storage and retrieval
systems and the users’ ability to effectively analyze and act on the information they
contain. Both relational and OLAP technologies have tremendous capabilities for
navigating massive data warehouses, but brute force navigation of data is not enough.
A new technological leap is needed to structure and prioritize information for specific
end-user problems. The data mining tools can make this leap. Quantifiable business
benefits have been proven through the integration of data mining with current
information systems, and new products are on the horizon that will bring this
integration to an even wider audience of users.

META Group Application Development Strategies: "Data Mining for Data

1

Warehouses: Uncovering Hidden Patterns.", 7/13/95 .

2
Gartner Group Advanced Technologies and Applications Research Note, 2/1/95.
3
Gartner Group High Performance Computing Research Note, 1/31/95.

Glossary of Data Mining Terms

analytical model A structure and process for analyzing a dataset. For example, a decision tree
is a model for the classification of a dataset.

anomalous data Data that result from errors (for example, data entry keying errors) or that
represent unusual events. Anomalous data should be examined carefully
because it may carry important information.

artificial neural Non-linear predictive models that learn through training and resemble
networks biological neural networks in structure.
CART Classification and Regression Trees. A decision tree technique used for
classification of a dataset. Provides a set of rules that you can apply to a new
(unclassified) dataset to predict which records will have a given outcome.
Segments a dataset by creating 2-way splits. Requires less data preparation
than CHAID.

CHAID Chi Square Automatic Interaction Detection. A decision tree technique used
for classification of a dataset. Provides a set of rules that you can apply to a
new (unclassified) dataset to predict which records will have a given
outcome. Segments a dataset by using chi square tests to create multi-way
splits. Preceded, and requires more data preparation than, CART.

classification The process of dividing a dataset into mutually exclusive groups such that
the members of each group are as "close" as possible to one another, and
different groups are as "far" as possible from one another, where distance is
measured with respect to specific variable(s) you are trying to predict. For
example, a typical classification problem is to divide a database of
companies into groups that are as homogeneous as possible with respect to a
creditworthiness variable with values "Good" and "Bad."

clustering The process of dividing a dataset into mutually exclusive groups such that
the members of each group are as "close" as possible to one another, and
different groups are as "far" as possible from one another, where distance is
measured with respect to all available variables.

data cleansing The process of ensuring that all values in a dataset are consistent and
correctly recorded.

data mining The extraction of hidden predictive information from large databases.

data navigation The process of viewing different dimensions, slices, and levels of detail of a
multidimensional database. See OLAP.

data The visual interpretation of complex relationships in multidimensional data.

visualization

data warehouse A system for storing and delivering massive quantities of data.

decision tree A tree-shaped structure that represents a set of decisions. These decisions
generate rules for the classification of a dataset. See CART and CHAID.

dimension In a flat or relational database, each field in a record represents a dimension.

In a multidimensional database, a dimension is a set of similar entities; for
example, a multidimensional sales database might include the dimensions
Product, Time, and City.
exploratory data The use of graphical and descriptive statistical techniques to learn about the
analysis structure of a dataset.

genetic Optimization techniques that use processes such as genetic combination,

algorithms mutation, and natural selection in a design based on the concepts of natural
evolution.

linear model An analytical model that assumes linear relationships in the coefficients of
the variables being studied.

linear regression A statistical technique used to find the best-fitting linear relationship
between a target (dependent) variable and its predictors (independent
variables).

logistic A linear regression that predicts the proportions of a categorical target

regression variable, such as type of customer, in a population.

multidimensional A database designed for on-line analytical processing. Structured as a

database multidimensional hypercube with one axis per dimension.

multiprocessor A computer that includes multiple processors connected by a network. See

computer parallel processing.

nearest neighbor A technique that classifies each record in a dataset based on a combination
of the classes of the k record(s) most similar to it in a historical dataset
(where k ³ 1). Sometimes called a k-nearest neighbor technique.

non-linear model An analytical model that does not assume linear relationships in the
coefficients of the variables being studied.

OLAP On-line analytical processing. Refers to array-oriented database applications

that allow users to view, navigate through, manipulate, and analyze
multidimensional databases.

outlier A data item whose value falls outside the bounds enclosing most of the other
corresponding values in the sample. May indicate anomalous data. Should
be examined carefully; may carry important information.

parallel The coordinated use of multiple processors to perform computational tasks.

processing Parallel processing can occur on a multiprocessor computer or on a network
of workstations or PCs.

predictive model A structure and process for predicting the values of specified variables in a
dataset.

prospective data Data analysis that predicts future trends, behaviors, or events based on
analysis historical data.

RAID Redundant Array of Inexpensive Disks. A technology for the efficient

parallel storage of data for high-performance computer systems.

retrospective Data analysis that provides insights into trends, behaviors, or events that
data analysis have already occurred.

rule induction The extraction of useful if-then rules from data based on statistical
significance.

SMP Symmetric multiprocessor. A type of multiprocessor computer in which

memory is shared among the processors.

terabyte One trillion bytes.

time series The analysis of a sequence of measurements made at specified time

analysis intervals. Time is usually the dominating dimension of the data.