Profit Chart (Analysis Services - Data Mining)
A profit chart displays the estimated profit increase that is associated with using a mining model. For example, if your model predicts which customers a company should contact in a business scenario, the profit chart incorporates information about the cost of conducting the targeted mailing campaign to contact x number of customers, and calculates the estimated profit. A typical profit chart shows an increase in profits up to a point, after which profits decrease as more of the population is contacted.
Understanding the Profit Chart
A profit chart is similar to a lift chart. Like a lift chart, a profit chart can be used to compare multiple models, as long as they all predict the same discrete attribute. There is no separate interface for creating a profit chart; you start by using Lift Chart tab of the Mining Accuracy Chart tab of Data Mining Designer, and then add in the cost and profit information that is specific to profit charts.
To illustrate how this works, this topic walks you through a profit chart that has been created to predict which prospective customers are more likely to buy a bike, and how much profit you might realize by selectively targeting these prospects.
To follow along in this scenario, use the decision tree model, TM_Decision Tree, which you created in the Basic Data Mining Tutorial. You start by selecting a model and predictable attribute, as you would for a lift chart, but and select Profit Chart from the list.
The Profit Chart Setting dialog box automatically opens whenever you choose profit chart as the chart type. This dialog box helps you specify the costs and benefits associated with a targeted mailing campaign. Once you have set the parameters that define a profit chart, the chart that is displayed automatically changes to a profit chart. For the chart shown in these examples, we used the following values:
Setting |
Value |
|---|---|
Choose a model |
TM_DecisionTree |
Set the predictable attribute and predictable value |
for this scenario, you really only want the customers who are likely to buy a bike, so choose [Bike Buyer] =1 In other scenarios, it might be more important to model negative costs: that is, you would want your profit chart to account for the cost of making a false prediction. In such a scenario, you would not specify any particular predictable value, and measure all outcomes. |
Choose the testing data set, or the data used to assess the accuracy and profitability of the model |
If you want to assess just the general accuracy of profitability of the model, you can use the test set that was generated when the mining structure was created. However, if you want to predict the accuracy and profitability of the model on actual data, use the data set that contains your prospective customers and their attributes. |
Set the value for the total target population |
Your database might contain many customers, but to save on mailing expenses you would target only the top 20,000 customers that the model identifies as most likely to respond. |
Enter the one-time cost of setting up a targeted mailing campaign for 20,000 people |
500 |
Enter the per-unit cost for the targeted mailing campaign. This amount will be multiplied by a number less than or equal to 20,000, depending on how many customers the model predicts are good prospects. |
3 |
Enter a value that represents the amount of profit or income that can be expected from a successful result. This amount will be used to project the total profit associated with high probability cases. |
25 |
Interpreting the Results
The following diagram shows the chart that was based on these parameters. The Y-axis of the chart represents the profit, while the X-axis represents the percentage of the population that the company contacted.
.gif "example of simple profit chart")
The profit chart contains a gray vertical line that marks a percentage of the target population. You can move the line by clicking a location in the chart. Each time you move the line, the Mining Legend is updated to display the percentage value, a profit score, and the predict probability that is associated with the population percentage at the vertical gray line. If you move the gray line to the point in the chart where profits are the highest, you can use the predict probability value to determine a strategy for contacting customers.
Percent Cases |
Series, Model |
Profit |
Predict Probability |
|---|---|---|---|
30 |
|
$103,000 |
67.23% |
40 |
TM_Decision Tree |
$128,500 |
60.90% |
50 |
|
$149,500 |
50.70% |
60 |
|
$168,000 |
44.05% |
By experimenting with this graph, you might determine that the peak of the profit curve is at 55 percent of the population and the associated predict probability is 20 percent. These results indicate that to achieve maximum profits you should only contact those customers whose response is predicted with a 20 percent or greater chance.
Related Content
The following topics contain more information about how you can build and use accuracy charts.
Topics |
Links |
|---|---|
Provides a walkthrough of how to create a lift chart for the Targeted Mailing model. |
Testing Accuracy with Lift Charts (Basic Data Mining Tutorial) |
Explains related chart types. |
Lift Chart (Analysis Services - Data Mining) |
Describes cross-validation for mining models and mining structures. |
|
Describes steps for creating lift charts and other accuracy charts. |