R语言RandomForest包Importance()输出中TRUE/FALSE列含义问询

Great question! I’ve stumbled on this exact confusion before when working with binary targets and binary features in the randomForest package—those columns labeled with your class levels (TRUE/FALSE) are easy to overlook in official docs, but they’re actually quite insightful.

What Are the TRUE/FALSE Columns? #

These columns represent class-specific Mean Decrease Accuracy values. Since your response variable is binary (with TRUE and FALSE as the two class levels), the importance() function automatically breaks down the overall MeanDecreaseAccuracy metric into how much each feature contributes to predicting each individual class.

How Are They Calculated? #

Let’s break down the logic clearly:

• The overall MeanDecreaseAccuracy measures the average drop in model accuracy when you permute (shuffle) a single feature’s values across all samples.
• The TRUE column: This is the average drop in accuracy only for samples belonging to the TRUE class when the feature is permuted. In short, it tells you how much worse the model gets at identifying TRUE cases when this feature is messed up.
• The FALSE column: Similarly, this is the average drop in accuracy only for samples belonging to the FALSE class when the feature is permuted.

This breakdown helps you spot which features are more critical for distinguishing one class over the other. For example, if a feature has a high value in the TRUE column but a low value in the FALSE column, it’s far more important for correctly classifying TRUE cases than FALSE ones.

How to Hide These Columns (If You Don’t Need Them) #

If you only want the standard overall metrics (MeanDecreaseAccuracy and MeanDecreaseGini), add the class=FALSE argument when calling importance():

importance(your_random_forest_model, class = FALSE)

内容的提问来源于stack exchange，提问作者AutoMiner