Scatter Plot (Analysis Services - Data Mining)
A scatter plot is automatically displayed in the Lift Chart tab if you choose any model that contains a continuous predictable attribute, other than a time series model. A scatter plot graphs the actual values in your data against the values predicted by the model. The scatter plot displays the actual values along the X-axis, and displays the predicted values along the Y-axis. It also displays a line that illustrates the perfect prediction, where the predicted value exactly matches the actual value. The distance of a point from this ideal 45-degree angle line indicates how well or how poorly the prediction performed.
This section explains how to create a scatter plot and how to interpret the results.
Only mining models that contain a continuous predictable attribute can be viewed in a scatter plot.
For example, consider a model in which the marketing department at Adventure Works Cycles predicts daily sales based on the number of clicks on a link sent in a promotional e-mail. Because both the number of clicks and the amount of sales are continuous numeric values, you can graph the number of clicks as the independent variable and the sales as the dependent variable. When you do so, the straight line shows the expected linear relationship, and the points scattered around that line show how the actual data diverges from the expected. This analysis tells you at a glance how closely a set of results is correlated with a particular input, and how much variation there is from the ideal model
The following diagram shows an example of a scatter plot, created for the scenario just described.
You can pause the mouse on any point scattered around the line to view the predicted and actual values in a tooltip. There is no Mining Legend for a scatter plot; however, the chart itself contains a legend that displays the score associated with the model. For more information about interpreting the score, see Mining Model Content for Linear Regression Models (Analysis Services - Data Mining).
You can copy the visual representation of the chart to the Clipboard, but not the underlying data or the formula. If you want to view the regression formula for the line, you can create a content query against the model. For more information, see Querying a Linear Regression Model (Analysis Services - Data Mining).
When you create a scatter plot, follow these steps:
In the Mining Accuracy Chart of Data Mining Designer, click the Input Selection tab.
In the Input Selection tab, select a model to evaluate. The model must contain a predictable attribute of a continuous numeric data type.
Select the predictable attribute.
Choose the data set to use in evaluation.
Optionally, apply a filter to the data set.
Click the Lift Chart tab to automatically generate a scatter plot report.
For a step-by-step procedure that applies to all chart types, see How to: Create an Accuracy Chart for a Mining Model.