Lift chart

A lift chart is used to evaluate the performance of a classification model.

A classification model, is simply a model which is used to predict a discrete variable.

An example lift chart is shown below.

Lift chart

A lift chart in Bayes Server can be used to evaluate the effectiveness of a model when predicting (classifying) a particular discrete state (e.g. Purchased=True). Consider the example of targeted marketing, where the purpose of your model is to select customers who are most likely to purchase your product. By ranking the predicted probability of a purchase, a chart can be generated that shows the effect of selecting cases with the highest probability first, as these are the cases that marketing would select based on the advice of your model. If the expected value for a case was indeed a purchase, then the chart will increase on the y axis (% of total purchases).

Three lines are displayed on the lift chart. Predicted, Ideal, and Random.

  • Predicted - displays the results for your model.
  • Ideal - displays the ideal scenario (no classification errors)
  • Random - displays the results based on a theoretical model which randomly selects cases.

By examining the chart, you can evaluate the performance of your model. In the targeted marketing scenario, you can also determine what percentage of cases should be chosen for marketing.

A lift chart can be generated from the Statistics tab in the Batch Query window. Both the expected value, and the predicted probability of a particular state are required.

The predicted probability of a state can be output using a PredicProbability query, e.g. PredictProbability(Purchased=True).

The expected value, can be included in the output, via an Information Column, defined in the Data map window.

An alternative way of accessing the expected value, is to map the predicted variable in the Data map window, and use retracted evidence so that the prediction does not use the expected value.