Lift Chart Tool
The Lift Chart tool produces a cumulative captured response chart (also called a gains chart) and an incremental response rate chart that are used to visually assess the comparative accuracy of different binary (yes/no) classification models to predict new data and to make an assessment of the expected economic implications of the use of a predictive model in a business process.
Both of these charts are based on aggregating data into 10 groups (deciles) that are ordered based on the predicted probability of a favorable response for each model, and then comparing this response to what would be expected if the selection of prospects was done randomly. In the case of the cumulative response chart, the chart examines what percentage of the total response that would be obtained if all customers in an organization's database were contacted is obtained by contacting the best 10 percent, 20 percent, and so on based on model predictions. The x (horizontal) axis of the chart is the percentage of the database contacted, while the y (vertical) axis is the percentage of the total response captured by going up to that "best" decile of the database based on the model.
For example, the best 20 percent of prospects based on the model may represent 50 percent of the favorable response that would be obtained if all prospects were contacted. The incremental response rate chart gives the favorable response rate for each of the model-based ordered decile groups in the database. In addition, the tool produces a gains table and measures of the area under the curve and the Gini coefficient to provide overall comparative metrics of the performance of different models when the cumulative captured response chart option is selected, and a table of response rates at different deciles for each model when the incremental response rate option is selected.
This tool uses the R tool. Go to Options > Download Predictive Tools and sign in to the Alteryx Downloads and Licenses portal to install R and the packages used by the R Tool. See Download and Use Predictive Tools.
The tool requires:
- Binary classification models, such as the Neural Network Tool, Logistic Regression Tool, Decision Tree Tool, Forest Model Tool, Boosted Model Tool, Spline Model Tool, or Stepwise Tool that have been unioned together into a single data stream.
- An Alteryx data stream or XDF metadata stream that is consistent with the model object (in terms of field names and the field types), and all have the same binary target variable.
- Lift chart type: Select the Total cumulative response chart or the Incremental response rate chart.
- True response rate (should be between 0 and 1): Often the database used to construct a predictive model has been "oversampled" to increase the percentage of observations that have one of the two target response categories. By specifying the true (original) response rate of the target variable, the lift chart constructed will account for the oversampling of one level of the target. If there is no oversampling, the value of this option should correspond to the response rate of the database used to develop the model(s).
- Target level (the label for the desired response of the target variable): This option allows for the creation of a lift chart that is consistent with the decision context, and allows for error checking to ensure that the chart created is relevant.
- Sample name (optional): This option allows the user to place a label on the created chart indicating what sample within the data stream was used to create the chart, or provide other information the data artisan views as relevant to the chart.
- The number of records to read at one time: Use to limit how many records are pulled in from the data stream at once.
- Graph resolution: Select the resolution of the graph in dots per inch: 1x (96 dpi); 2x (192 dpi); or 3x (288 dpi). Lower resolution creates a smaller file and is best for viewing on a monitor. Higher resolution creates a larger file with better print quality.
View the output
The output for the tool is an Alteryx Report field that consists of an R-Graph and a Lift/Gains table. These elements can be used to assist in the creation of custom reports.