I’m working on a ML project and I’m trying to train some classification models and then make some predictions on the test df.

How do I train a model that’s capable of using every observation available, regardless it has missing values or not? And how do I make predictions?

Most of the observations from my df have at least one missing value.
To train my models I’m using the library caret.

For example, given this model:

control <- trainControl(method="cv", number=10, search="grid", summaryFunction = twoClassSummary, classProbs = TRUE)
tunegrid <- expand.grid(.mtry=c(1:6))
rf <- train(Target~., data=train, method="rf", metric="ROC", tuneGrid=tunegrid, ntree=100,trControl=control)

and then I make predictions this way:

test$pred<-predict(rf,test,'prob')[,2]

While training, I’ve already tried this na.action options:

• na.omit;

–na.exclude;

• na.pass.

The first two work fine, but if I use na.pass I get this error:

Something is wrong; all the ROC metric values are missing:
      ROC           Sens          Spec    
 Min.   : NA   Min.   : NA   Min.   : NA  
 1st Qu.: NA   1st Qu.: NA   1st Qu.: NA  
 Median : NA   Median : NA   Median : NA  
 Mean   :NaN   Mean   :NaN   Mean   :NaN  
 3rd Qu.: NA   3rd Qu.: NA   3rd Qu.: NA  
 Max.   : NA   Max.   : NA   Max.   : NA  
 NA's   :6     NA's   :6     NA's   :6    
Error: Stopping
warnings()
1: model fit failed for Fold01: mtry=1 Error in randomForest.default(x, y, mtry = param$mtry, ...) : 
  NA not permitted in predictors

And if I use one of the first two, when I make predictions, I get erorrs similar to this one:

Error in set(x, j = name, value = value) : 
  Supplied 199 items to be assigned to 5425 item