REVIEWER: Levi A. Sorrell, PhD, MBA, director of advanced analytics at Janssen Pharmaceutica.

BACKGROUND: Janssen Pharmaceutica is a leading developer and manufacturer of prescription drugs and is part of the pharmaceutical sector of Johnson & Johnson.

PLATFORMS: WizWhy is currently deployed on Compaq AP 500s running dual Pentium II 450s under NT 4 sp5 with 256MB of memory, and SP 700s running dual 500 Xeons under Windows 2000 with 512MB of memory.

PROBLEM SOLVED: We use WizWhy in two ways: to validate the contents of new data sets when they arrive and to narrow the scope of variables included in more traditional analyses. WizWhy is used to analyze promotional effectiveness, sales rep effectiveness and response patterns of highly detailed consumer surveys of behavior.

PRODUCT FUNCTIONALITY: Rule-generation tools are particularly good at giving the user interesting findings. Revealing interesting phenomena relies on the assumption that unexpected phenomena are interesting. For example, an event that is inconsistent with (unexpected by) an accepted theory is an interesting event. The one-condition rules and trends can be viewed as the theory that describes the data. By calculating how unlikely each rule is relative to the basic trends, the unexpected rules are revealed. These unexpected rules signify the interesting phenomena in the data.

STRENGTHS: WizWhy has several strengths that are unique. Among these strengths are two new capabilities in version 3: the if-and-only-if rules, and the use of multivalue-dependent variables. If-and-only-if rules are sets of necessary and sufficient conditions such that if at least one of the conditions holds, the dependent variable has a certain value in, for instance, 90 percent of the cases. If all of the conditions do not hold, then the dependent variable does not have this value in, for instance, 95 percent of the cases. Version 3 also supports the use of multivalue-dependent variables as well as Boolean dependent variables. When the dependent is categorical, WizWhy reveals rule and issue predictions in regard to each one of the dependent variable's values. When the dependent is continuous, WizWhy divides it into nine intervals and reveals the rule and issue predictions in regard to each of these intervals.

WEAKNESSES: Its ease of use strength is also its weakness. While WizWhy works hard to warn against overfitting your data, overfitting is a subtlety that neophyte data miners are likely to miss. WizWhy attempts to handle overfitting and rules based on noise in the data by generating the error probability of a rule and the unlikelihood of the rule. This may not be enough. By giving the semblance of ease, WizWhy may lull users into a false sense of expertise.

SELECTION CRITERIA: The major reasons we chose WizWhy were the relative longevity of the company, its ease of use, its connectivity and access to the brainpower of its creators. It probably should not go without saying that we were first and foremost happy with its technical finesse.

DELIVERABLES: The product delivers all the if-then rules discovered in our data, as well as the necessary and sufficient conditions (if-and-only-if rules), error probability of each rule, best segmentations of the continuous fields, predictive power of each field and unexpected rules it found. It summarizes the data graphically by presenting the main rules and trends. It can be used to predict new cases on the basis of the discovered rules using error costs and to explain those predictions by listing the relevant rules that were applied. These unexpected cases might be the result of noise, data-entry errors or fraud.

VENDOR SUPPORT: Vendor support is outstanding and includes direct conversations with the president of the company. Recommendations for changes and improvements are given serious consideration. The company responds with modifications, a timetable for changes or reasons why the change cannot be made at this time.

DOCUMENTATION: The documentation was quite useful right out of the box for experienced data miners and statisticians. The introductory material on rule-generation tools and the examples provided allow anyone with an analytical background to begin using WizWhy immediately. Some exposure to data mining is needed to ensure the tool is used to answer well conceived questions, although rule-generation tools tend to be straightforward. While the manuals are good guides for experienced analysts, inexperienced users will find them too technical in their focus.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access