User manual BUSINESS OBJECTS PREDICTIVE ANALYSIS 6.5.1

[. . . ] Predictive Analysis User's Guide Predictive Analysis 6. 5. 1 2 Predictive Analysis User's Guide Copyright Copyright © 2004 Business Objects. All rights reserved. Trademarks Business Objects, the Business Objects logo, Crystal Reports, and Crystal Enterprise are trademarks or registered trademarks of Business Objects S. A. All other names mentioned herein may be trademarks of their respective owners. This software and documentation is commercial computer software under Federal Acquisition regulations, and is provided only under the Restricted Rights of the Federal Acquisition Regulations applicable to commercial computer software provided at private expense. [. . . ] - Independent: Model refresh is scheduled independently of set refresh - for example, through the scheduler or a rule. The scope of model refresh can be two-fold: · Refresh Statistics only: this causes the refresh of most statistics and metrics associated with the model whenever the model is refreshed. · Regenerate model: this not only refreshes the statistics, but also re-generates the optimal binnings and recompute the formulas used in generating model scores (exposed in Individual List analytic). A few statistics are also calculated only at the time of model regeneration - Net Relevance, in particular. NOTE Regenerating a model requires much more computation time than does statistics only refresh, as it necessitates multiple passes through the data. 8. Click Finish. NOTE To manually regenerate or refresh a model click on the links. Model based metrics Models may be thought of as batch calculation engines for statistics. These statistics are exposed in Predictive Analysis as well as in time series analyticssuch as Interactive Analytic. Typically, a statistic calculated by a model quantifies a relationship between an influencer and a goal variables but it can also be a descriptive statistic that summarizes some aspect of the influencer's distribution in the population with which the model is associated (for example, the mean of Age or the max of Purchase Size in Last 30 Days). Configuring Predictive Analytics Predictive Analysis User's Guide 33 Certain statistics can be valuable for automating model management, such as Root Mean Squared Error or goal variable Obsolescence. These can be used in rules to detect when it is time to regenerate a model in statistics-only refresh mode or to trigger model regeneration when one whose refresh is unscheduled (for example, used for scoring lists, but not for statistics calculation). NOTE Model based metrics does not roll-up in the past as you can get only one value based on the last model refresh information. Click Add and select the New model based metric. . Models 34 Predictive Analysis User's Guide The Create Metric window is displayed. 2. Select the Aggregation for the model and the influencer. NOTE Refer to table on page 36 for available aggregations. These aggregations are correspond to either model, influencer or bin level statistics that you select. Click Next to now create the attributes. Configuring Predictive Analytics Predictive Analysis User's Guide 35 On the attribute window you have to define the various attribute to the new metric. The duration of the time periods or slices that make up a metric is called the grain. The metric grain can vary from small spans of time to longer and is dependent on the calendar -- the finer the grain, the shorter the time period between metric value calculations. Refresh Type allows you to specify on what basis want your metric refreshed. It can be: · With Model: this refreshes the new metric as and when the model is refreshed. · Independent: this refreshes the metric independent of any other parameters used by scheduler and rules. · External Refresh Storage options Because metric values require very little storage, it usually makes sense to store the full metric history so that trending can be done. You can, however, also choose only to store the most recent value or disable the storage. Parameters The Default Smoothing list allows you to specify the statistical transformation that is used in reporting on the metric when the user accepts the Default Smoothing selection in the Smoothing list. The nature of the data you are probing and which questions you hope to answer determine the type of smoothing that is most suitable. Available transformations are determined by your administrator during installation and configuration. [. . . ] If the Derived Variable displays properly, it's syntax is correct. Once you've tested each Derived Variable, you're ready to move onto Model Building. NOTE The query that is run for the model must always return a single row per key. Because predictive models cannot do runtime query aggregation, the data used with the metrics universe must already be summarized to the level of one row per key value. To test derived variables 72 Application Foundation User's Guide Model building To build a model some tips to follow: 1. Create a simple model, using a small population (<1000 rows), one Influencer and one Goal. [. . . ]