Case Study: Local Explanations with ICE Curves
Learn how to apply ICE plots to explore how the instance’s prediction changes when a feature changes.
In this lesson, we’ll learn about a technique called individual conditional expectation (ICE) curves, which helps us understand how changing a specific feature affects the predictions of a machine learning model for each individual data point.
Individual conditional expectation (ICE) curves show how the prediction of each individual data point changes when a feature is changed. Each data point is represented by a line on the ICE curve, illustrating how the prediction varies with different feature values.
What is an ICE curve?
An ICE curve is a way to show how changing one specific feature in a machine learning model affects the predictions for individual data points. It does this by plotting a separate line or curve for each data point, which shows how the model’s output changes when that feature is changed while keeping the other features constant. ICE plots give us detailed insights into how the model behaves for each individual data point.
An ICE plot is a more detailed version of a partial dependence plot (PDP). PDPs give us a summary chart that shows how changing one feature affects the model’s predictions on average. They do this by plotting the average predicted outcomes for different values of the feature we’re interested in while keeping all other features constant.
However, PDPs can ...