Dela via


How to: Create an Accuracy Chart for a Mining Model

You can create an accuracy chart in SQL Server Analysis Services in five basic steps:

  • Select the mining structure that contains the mining models that you want to compare.

  • Select the mining models to add to the chart.

  • Specify a source of testing data to use in generating the chart.

  • Choose the chart type.

  • Configure the chart options.

These basic steps are the same for the lift chart, profit chart, and classification matrix. The following procedures outline the steps to configure the basic chart options for these chart types. For information about how to create a cross-validation report, see Cross-Validation Report (Analysis Services - Data Mining).

To select a mining structure

  1. Open the Data Mining Designer in Business Intelligence Development Studio.

  2. In Solution Explorer, double-click the structure that contains the mining model or models.

  3. Click the Mining Accuracy Chart tab.

To select mining models for inclusion in the chart

  1. On the Mining Accuracy Chart tab of Data Mining Designer in Business Intelligence Development Studio, click the Input Selection tab.

    The list displays all models in the current structure that have the same predictable attribute.

  2. Select the Show box for each model that you want to include in the chart.

  3. Click the Predictable Column Name text box, and select the name of a predictable column from the list. All models that you put in one chart must have the same predictable column.

  4. If you compare two models and the predictable columns have different values or different data types, clear the Synchonize prediction columns and values box to force a comparison.

    Note

    If the Synchonize prediction columns and values box is selected, Analysis Services analyzes the data in the predictable columns of the model and the test data, and attempts to find the best match. Therefore, do not clear the box unless absolutely necessary to force a comparison of the columns.

  5. Click the Predict Value text box, and select a value from the list. If the predictable column is a continuous data type, you must type a value in the text box.

    For more information, see How to: Select a Predictable Column for an Accuracy Chart.

To select testing data

  • On the Input Selection tab of the Mining Accuracy Chart tab, specify the source of the data that you will use to generate the chart by selecting one of the options in the group, Select data set to be used for accuracy chart.

    1. Use Mining Model test cases   Select the option.

    2. Use mining structure test cases   Select the option.

    3. Specify a different data set   Select the option, and then click the browse () button to choose the external data tables to use for the accuracy chart. For more information, see How to: Select Input Data for an Accuracy Chart and How to: Modify Column Mappings in an Accuracy Chart.

      If you are using an external data set, you can optionally filter the input data set. For more information, see How to: Filter the Input Rows for an Accuracy Chart.

Note

You cannot create a filter on the model test cases or the mining structure test cases on the Input Selection tab. To create a filter on the mining model, modify the Filter property of the model. For more information, see How to: Apply a Filter to a Mining Model.

To generate the chart and configure optional chart settings

  1. In the Mining Accuracy Chart tab, click the tab for the chart you want to create.

  2. For a lift chart, click the Lift Chart tab.

    The chart is automatically generated based on the selections you made in the Input Selection tab. For more information about how to use lift charts, see Lift Chart (Analysis Services - Data Mining).

  3. For a profit chart, first click the Lift Chart tab. From the Chart type drop-down list, select Profit chart.

    The Profit Chart Settings dialog box opens. For more information about how to configure profit chart options, see Profit Chart (Analysis Services - Data Mining).