About the missing values in the feature matrix

Question

[Eric Wang]

Hi all,

I am studying the classification problem (or prediction problem) based on a feature matrix. But I found a lot of missing values in some features (even more than 80%). I would like to ask how to add or fill in the missing values while retaining these features. At the same time, may I ask if such a filling is meaningful? The reason why I want to keep these features is that after I remove the missing values, the classification results based on individual features are good.

Best Regards

Eric

Answer 1

[Mensur Dlakic]

You do not need to fill in (impute) the missing values. What you have is a sparse matrix, and plenty of classification tools work with sparse data. Just to name a few: support vector machines, random forests, gradient boosting machines.