Filling missing values in the data with approximating values. Common choices include the expected value (mean), most common value for categoricial features or using an ML model to predict the most likely values given some other features in the data.