MicroStrategy ONE

Aggregating Predictive Metrics

One of the most powerful features of integrating data mining models with a business intelligence platform is the ability to drill up and down through layers of data. For predictive results to make sense, the aggregation function to be used must be specified.

The steps that need to be taken to ensure the predictive metric aggregates properly under all circumstances are described below:

  • Place the predictive function within an aggregation function.
  • Specify the dynamic aggregation function on the predictive metric. This is required only when the predictive metric is used with MicroStrategy OLAP Services.

Choosing proper aggregation function requires some knowledge about how the model behaves. For example:

  • If the predictive metric generates a score that is a zero or a one, use Sum to calculate the number of "one" scores.
  • If the predictive metric generates a "linear" output, like "Forecasted Revenue," usually from a regression predictor, use Sum to roll up the predictive results.
  • If the predictive metric generates a confidence or percentage, use Average to calculate the mean confidence.
  • If the predictive metric generates a numeric classifier, like a cluster/segment number, use Mode to calculate the most common classifier.
  • For models with outputs that cannot be aggregated, select "None" as the aggregation function.

While an explicit aggregation function is useful for normal reports, deployments that take advantage of MicroStrategy OLAP Services should also set the predictive metric's dynamic aggregation function. The Import Data Mining Model feature in MicroStrategy Developer does this automatically, but you can also set the dynamic aggregation function using the Metric Editor.

To Set the Dynamic Aggregation Function

  1. Open the predictive metric in the Metric Editor.
  2. Select the Subtotals/Aggregation tab.
  3. Select the appropriate Dynamic Aggregation function.