Evaluation Plot Tab

Chart type. Select one of the following types: Gains, Response, Lift, Profit, ROI (return on investment), or ROC (receiver operator characteristic).

Cumulative plot. Select to create a cumulative chart. Values in cumulative charts are plotted for each quantile plus all higher quantiles. (Cumulative plot is not available for ROC charts.)

Include baseline. Select to include a baseline in the plot, indicating a perfectly random distribution of hits where confidence becomes irrelevant. (Include baseline is not available for Profit and ROI charts.)

Include best line. Select to include a best line in the plot, indicating perfect confidence (where hits = 100% of cases). (Include best line) is not available for ROC charts.)

Use profit criteria for all chart types. Select to use the profit criteria (cost, revenue, and weight) when calculating the evaluation measures, instead of the normal count of hits. For models with certain numeric targets, such as a model that predicts the revenue obtained from a customer in response to an offer, the value of the target field gives a better measure of the performance of the model than the count of hits. Selecting this option enables the Costs, Revenue, and Weight fields for Gains, Response, and Lift charts. To use the profit criteria for these three chart types, it is recommended that you set Revenue to be the target field, Cost to be 0.0 so that profit is equal to revenue, and that you specify a user defined hit condition of "true" so that all records are counted as hits. (Use profit criteria for all chart types is not available for ROC charts.)

Find predicted/predictor fields using. Select either Model output field metadata to search for the predicted fields in the graph using their metadata, or select Field name format to search for them by name.

Plot score fields. Select this check box to enable the score fields chooser. Then select one or more range, or continuous, score fields; that is, fields which are not strictly predictive models but which might be useful to rank records in terms of propensity to be a hit. The Evaluation node can compare any combination of one or more score fields with one or more predictive models. A typical example might be to compare several RFM fields with your best predictive model.

Target. Select the target field using the field chooser. Choose any instantiated flag or nominal field with two or more values.

Note: This target field is only applicable to score fields (predictive models define their own targets), and is ignored if a custom hit criterion is set on the Options tab.

Split by partition. If a partition field is used to split records into training, test, and validation samples, select this option to display a separate evaluation chart for each partition. See the topic Partition Node for more information.

Note: When splitting by partition, records with null values in the partition field are excluded from the evaluation. This will never be an issue if a Partition node is used, since Partition nodes do not generate null values.

Plot. Select the size of quantiles to plot in the chart from the drop-down list. Options include Quartiles, Quintiles, Deciles, Vingtiles, Percentiles, and 1000-tiles. (Plot is not available for ROC charts.)

Style. Select Line or Point.

For all chart types except ROC charts, additional controls enable you to specify costs, revenue, and weights.

• Costs. Specify the cost associated with each record. You can select Fixed or Variable costs. For fixed costs, specify the cost value. For variable costs, click the Field Chooser button to select a field as the cost field. (Costs is not available for ROC charts.)

• Revenue. Specify the revenue associated with each record that represents a hit. You can select Fixed or Variable costs. For fixed revenue, specify the revenue value. For variable revenue, click the Field Chooser button to select a field as the revenue field. (Revenue is not available for ROC charts.)
• Weight. If the records in your data represent more than one unit, you can use frequency weights to adjust the results. Specify the weight associated with each record, using Fixed or Variable weights. For fixed weights, specify the weight value (the number of units per record). For variable weights, click the Field Chooser button to select a field as the weight field. (Weight is not available for ROC charts.)