About Data Mining

  To mine an Essbase database you perform the following steps:

1. Build a data mining model.

   1. Select an algorithm to use.  

      Select one of the built-in algorithms provided by Data Mining Framework or add and select an algorithm from a third-party vendor. The algorithm determines how the data is searched. You choose an algorithm based on the business case or problem that you need to analyze.

   2. Specify data to train (build) the model.

      Write expressions to specify data of interest in the Essbase database. The specified data serves as training data for the algorithm.

   3. Train the model.

      Run the algorithm using the training data. During the training process, the algorithm discovers and describes patterns and relationships in the data that can be used for prediction. Later, the trained model can use the patterns and relationships to generate new information from a different, but similarly structured, set of data.

2. Optional: Test the model.

   Apply the model to a new set of data with known results or correlations in the database. By comparing the results generated by the algorithm to the known results, you can determine the accuracy of the algorithm.

3. Use the model to mine the data using either method:

   1. Apply the model.

      Applying the model writes results to the database:

      1. Select a trained model to use.

      2. Change the expressions that define the data to use. For example, if the model is trained on first quarter sales, you can apply the model to predict second and third quarter sales.

      3. Run the model to generate data mining results.

   2. Score the model.

      Scoring the model returns results interactively. Results are not written to the database.

      1. Select a scoring method and a trained model to use.

      2. Change the expressions that define the data to use.

         • If scoring on cube, specify data within the database.

         • If scoring on data, type in data values.

      3. Run the model to generate data mining results.