Skip to main content

Task 9: ER: LIFT

Continuing on the diagnostics page, select the LIFT curve. The Lift curve should look similar to the one below:

diagnostics-lift

note

A Lift chart visually assesses model performance.

  • Lift measures how effective a predictive model is by comparing its results to those without the model.
  • It's calculated by comparing the model's predicted results to results without any model.
  • A larger area between the lift curve and the baseline indicates a better model.
  • The lift chart shows how much more often positive instances appear in the model's top predictions compared to random selection. It answers: "For the top 1%, 2%, 10%, etc., of the model's predictions, how frequently are positive instances found?" The lift value reaches 1.0 at the 100th percentile, meaning all positive instances are captured by the model at that point.

The y-axis of the plot has been adjusted to represent quantiles; this allows for focus on the quantiles that have the most data and, therefore, the most impact.

  1. Hover over the various quantile points on the Lift chart to view the quantile percentage and cumulative lift values.

  2. What is the cumulative lift at 1%, 2%, 10% quantiles? diagnostics-lift-10-percent

    The lift chart organizes model predictions from highest to lowest scores, showing more uncertainty as we move along the x-axis (higher quantiles). For instance, at the 10th percentile, the model predicts a lift of 5.3. This means the top 10% of predictions include defaults at a rate 5.3 times higher than the baseline.

  3. We can judge how effective the model is based on the area between the lift curve and the baseline (white horizontal dashed line). What does a good lift curve look like?

The area between the baseline (white horizontal dashed line) and the lift curve (yellow curve) better known as the area under the curve visually shows us how much better our model is than that of the random model.

  1. Exit out of the Lift chart by clicking on the X located at the top-right corner of the plot, next to the DOWNLOAD option
Feedback