Random forest model in R - predictors and training data types mismatch

Question

I tried the random forest model in my research topic, but I met a problem during the validation phase.

When, I used the final model of random forest to predict on an independent dataset, I received this message:

Type of predictors in new data do not match that of the training data

So, to detect the different categories in my factors/variables, I used:

levels(Train$Aquifer.media)           levels(Test$Aquifer.media)

For this factor "Aquifer.media", I have:

Train dataset: "Carbonates rocks"  "Crystalline rocks"  "Siliciclastic sedimentary rocks"  "Unconsolisated sediments rocks"  "Volcanic rocks"

Test Dataset: "Crystalline rocks"  "Siliciclastic sedimentary rocks"  "Unconsolisated sediments rocks"  "Volcanic rocks"

I detected that predictors were of different categories, I would like to know, how I can solve this problem?

Is it possible to delete some categories in the factors?

Answer 1

Your training set should be true representation of the entire population which is not true in your case. The levels in your train data set's media column has 4 factor levels which is 1 level less than the test data set's media column factor levels. Assuming you are using R, you can fix it with below code

levels(TrainAquifier.media) <- levels(TestAquifier.media)

You can find answer to similar question in stackoverflow here