Solving Operational Business Intelligence with InfoSphere Warehouse Advanced Edition

368 Solving Operational Business Intelligence with InfoSphere Warehouse Advanced Edition

parental status, and so on. The system can then use predictive models

(

classification) to generate a credit risk score in real time that can be used as a

basis for either denying the application outright, or recommending for approval.

The scoring might also be used to indicate a best fit interest rate or repayment

terms, depending on risk level. The process might be more nuanced in that the

system makes a recommendation that is then followed up by a loan agent. The

process might also be used for over-the-phone applications or account changes.

The process for designing, training, validating, and deploying the predictive data

mining model in this scenario is the same as in the case of credit card fraud

previously described. The scoring method must be integrated with or otherwise

invokable from the online application or help-desk application, wherever the

business process is being driven, either by the consumer or company

representative. We discuss how these functions can be deployed for application

integration in subsequent sections in this chapter.

10.2 InfoSphere Warehouse Advanced Enterprise

Edition 10.1 tools and features for data mining

InfoSphere Warehouse Advanced Enterprise Edition 10.1 provides all of the

tools and features required to develop, train, validate, visualize, deploy, and

integrate data mining applications into an operational warehouse business

intelligence solution. Through the rest of this section, we review the InfoSphere

Warehouse features available to perform

򐂰 Source data exploration

򐂰 Data preparation

򐂰 Modeling in Design Studio

򐂰 Model validation and execution

򐂰 Visualization

򐂰 Scoring with other models through PMML

The data mining features of InfoSphere Warehouse is thoroughly described in

other IBM Redbooks publications, most notably InfoSphere Warehouse: A

Robust Infrastructure for Business Intelligence, SG24-7813. This section draws

heavily from that document at a summary level. For more detailed explanations

and examples, refer directly to InfoSphere Warehouse: A Robust Infrastructure

for Business Intelligence.

10.2.1 Source data exploration in InfoSphere Warehouse 10.1

After the business problem is identified and understood, and the data mining

method is selected to solve the problem, the next step in the process is source

Chapter 10. Techniques for data mining in an operational warehouse 369

data exploration. It is critical to understand where the input data is coming from,

how it is structured, and its granularity and relationship to other source data.

InfoSphere Warehouse 10.1 Design Studio has powerful features to explore and

profile the source data before modeling begins.

Database enablement

Before anything related to data mining can take place, the DB2 database must

be enabled for data mining. This is done by right-clicking the database name in

the Design Studio Data Source Explorer and selecting Enable Database for

Data Mining, as shown in Figure 10-1.

After DB2 is enabled, all data mining and data exploration function can be used.

Figure 10-1 Enabling DB2 for data mining

Source data exploration

InfoSphere Warehouse 10.1 Design Studio offers several functions to aid the

data mining modeling in the process of exploring the source data under

consideration.

370 Solving Operational Business Intelligence with InfoSphere Warehouse Advanced Edition

Browsing and sampling the data

In the Data Source Explorer, you can right-click a table and select to return all

rows of the table (with all columns) or to a sample set of the rows (with all

columns); see Figure 10-2.

Figure 10-2 Simple data exploration techniques in Design Studio

Figure 10-3 shows a sample result set.

Figure 10-3 Sample data results in Design Studio

Chapter 10. Techniques for data mining in an operational warehouse 371

Multi-table exploration

You can use general purpose SQL queries to browse and explore data in the

source tables using Design Studio. Right-click the database and select New SQL

Script, as shown in Figure 10-4.

Figure 10-4 Using general-purpose SQL in Design Studio

This operation opens a script editor in which one or more semi-colon-delimited (;)

SQL statements can be entered. The results are viewed in the SQL Results tab

just like the results shown in Figure 10-3 on page 370.

Record value distribution analysis techniques

InfoSphere Warehouse 10.1 Design Studio also contains more sophisticated

data exploration techniques that use statistical analysis functions. The analysis

functions investigate value distributions on a column-by-column basis, and

illuminate relationships between values of columns. The following analyses are

available:

򐂰 Univariate analysis

Univariate analysis shows both graphical and detailed value distribution

statistics for every column in the table. You can drill into the details for any

given column for information regarding the loyalty program participation for an

individual customer (from the Customer Insight example). See Figure 10-5 on