Portals eNewsletters Web Seminars dataWarehouse.com DM Review Magazine
DM Review | Covering Business Intelligence, Integration & Analytics
   Covering Business Intelligence, Integration & Analytics Advanced Search
advertisement

RESOURCE PORTALS
View all Portals

WEB SEMINARS
Scheduled Events

RESEARCH VAULT
White Paper Library
Research Papers

CAREERZONE
View Job Listings
Post a job

Advertisement

INFORMATION CENTER
DM Review Home
Newsletters
Current Magazine Issue
Magazine Archives
Online Columnists
Ask the Experts
Industry News
Search DM Review

GENERAL RESOURCES
Bookstore
Buyer's Guide
Glossary
Industry Events Calendar
Monthly Product Guides
Software Demo Lab
Vendor Listings

DM REVIEW
About Us
Press Releases
Awards
Advertising/Media Kit
Reprints
Magazine Subscriptions
Editorial Calendar
Contact Us
Customer Service

Data Warehousing Lessons Learned:
Data Mining is Dead Long Live Predictive Analytics!

  Column published in DM Review Magazine
January 2004 Issue
 
  By Lou Agosta

This is why data mining is dead: it died of a broken heart. It was killed by disappointed expectations. In addition to a perfect storm of tough economic times, another reason data mining technology has not lived up to its promise is that "data mining" is a vague and ambiguous term. It overlaps with data profiling, data warehousing, and even such approaches to data analysis as online analytical processing (OLAP) and enterprise analytic applications. When high profile successes have occurred (e.g., a front-page article in the Wall Street Journal, "Lucky Numbers: Casino Chain Mines Data on Its Gamblers, And Strikes Pay Dirt" by Christina Binkley, May 4, 2000), they have been a mixed blessing. Such results have attracted a variety of imitators with claims, solutions and products that ultimately fall short of the promises. The promises build on the mining metaphor and typically are made to sound like easy money. This has resulted in all the usual dilemmas of confused messages from vendors, hyperbole in the press and disappointed end-user enterprises.

Data mining is regrouping as "predictive analytics." The differentiators are summarized in Figure 1.


Figure 1: Data Mining and Predictive Analytic Differentiators

  • Prescriptive, not merely descriptive: Scanning through a terabyte haystack of billing data for a few needles of billing errors is properly described as data mining. However, it is descriptive, not prescriptive. When a model is able to predict errors based on a correlation of variables ("root cause analysis"), then the analysis is able to recommend what one ought to do about the problem (and is, therefore, prescriptive). Note that the model expresses a "correlation," not a "causation," though a cause-and-effect relation can often be inferred. For example, Xerox uses Oracle's Data Mining software, for clustering defects and building predictive models, to analyze usage profile history, maintenance data and representation of knowledge from field engineers to predict photocopy component failure. The copier then sends an e-mail to the repair staff to schedule maintenance prior to the breakdown.
  • Stop predicting the past; predict the future: Market trend analysis as performed in data warehousing, OLAP and analytic applications often asks what customers are buying or using (product or service), and then draws a straight line from the past into the future, extrapolating a trend. This too can be described as data mining. One might argue this predicts the future because it says something about what will happen. However, a more accurate description would be that it "predicts the past" and then projects that into the future. The prediction is not really in the analysis. Furthermore, data mining in the limited sense used here is only able to envision continuous change -- extending the trend from past to future. Predictive analytics is also able to generate scores from models that envision discontinuous changes -- not only peaks and valleys, but cliffs and crevasses. This is especially the case with "black box"-type functions such as neural networks and genetic programming. Rarely do applications in OLAP, query and reporting or data warehousing explicitly relate independent and dependent variables, but that is of the essence in predictive analytics. For example, KXEN is used to find the optimal point between savings of catching a bad customer versus the cost of turning away a good paying customer (opportunity cost).
  • Invent hypotheses, don't merely test them: Finally, data mining is distinguished from predictive analytics in terms of hypothesis formulation and validation. For example, one hypothesis is that people default on loans due to high debt. Once the analyst formulates this hypothesis by means of imaginative invention out of her or his own mind, the OLAP analyst then launches queries against the data cube to confirm or invalidate this hypothesis. Predictive analytics is different in that it can look for patterns in the data that are useful in formulating hypotheses. The analyst might not have thought that age was a determinant of risk, but a pattern in the data indicates that as a useful hypothesis for further investigation.

One reason that data alone is not knowledge but merely data is that it lacks structure, organization, direction, coherence, point and conceptual focus. Just as a predictive model without supporting data would be empty, likewise data without a unifying model is meaningless and leaves the collector blind. Giga clients will need to have expertise in all three dimensions: details of the business, data collection and model building. Client predictive efforts should be guided by the methodological injunction that determining meaning is a business task, not a statistical one. Within such a context, the selection of a tool for predictive analytics can be leveraged to the advantage of customer recommendations, cross- selling, up-selling, personalization, loyalty development, attrition and churn (reduction), forecasting, demand planning, inventory (and cost) reduction, brand development and the mastery of market dynamics.

...............................................................................

For more information on related topics visit the following related portals...
Data Mining.

Lou Agosta, Ph.D., joined IBM WorldWide Business Intelligence Solutions in August 2005 as a BI strategist focusing on competitive dynamics. He is a former industry analyst with Giga Information Group, has served as an enterprise consultant with Greenbrier & Russel and has worked in the trenches as a database administrator in prior careers. His book The Essential Guide to Data Warehousing is published by Prentice Hall. Agosta may be reached at LoAgosta@us.ibm.com.

Solutions Marketplace
Provided by IndustryBrains

Autotask: The IT Business Solution
Run your tech support, IT projects and more with our web-based business management. Optimizes resources and tracks billable project and service work. Get a demo via the web, then try it free with sample data. Click here for your FREE WHITE PAPER!

Life Cycle Management - Share Cad With Everyone
Share Cad projects with non-tech users. Get everyone on the same page. Request free information.

Design Databases with ER/Studio: Free Trial
ER/Studio delivers next-generation data modeling. Multiple, distinct physical models based on a single logical model give you the tools you need to manage complex database environments and critical metadata in an intuitive user interface.

Validate Data at Entry. Free Trial of Web Tools
Protect against fraud, waste and excess marketing costs by cleaning your customer database of inaccurate, incomplete or undeliverable addresses. Add on phone check, name parsing and geo-coding as needed. FREE trial of Data Quality dev tools here.

Grid Computing: Cut through the hype
Free Forrester Report discusses developing trends in grid technologies, breaks down the three major ways grid is being used and provides survey results from organizations that are piloting, rolling out, or currently leveraging grid technology.

Click here to advertise in this space


View Full Issue View Full Magazine Issue
E-mail This Column E-Mail This Column
Printer Friendly Version Printer-Friendly Version
Related Content Related Content
Request Reprints Request Reprints
Advertisement
advertisement
Site Map Terms of Use Privacy Policy
SourceMedia (c) 2006 DM Review and SourceMedia, Inc. All rights reserved.
SourceMedia is an Investcorp company.
Use, duplication, or sale of this service, or data contained herein, is strictly prohibited.