how to deal with NA in neural network prediction result in R

Question

I'm applying neuralnet on titanic dataset (containing PClass, sex, Age, Sibsp, Parch, Fare, Embarked)

library(caret)
model_nnet <- train(as.factor(Survived) ~.,  
              method="nnet",
              train_df, 
              linout=FALSE, 
              trace = FALSE,
              preProcess = c("center", "scale"))

nnet_predict <- predict(model_nnet, test_df)

While I expected nnet_predict to be same length as testing dataframe (418 records), it actually contains NA and only have 331 results. Any advice on how to deal with it? Thank you

1.618 1.618 · Accepted Answer · 2017-08-05T20:13:14

Look for

summary(test_df)

You can see that there are missing values in Age & Fare column so before running predict() function you need to fix NA in these two columns.

One option could be to -

Fill NA in Fare column with it's mean value.
Fill NA in Age column with it's mean value wrt Pclass i.e.

if Pclass==1 then missing_age <- 37
if Pclass==2 then missing_age <- 29
else missing_age <- 24

Hope this helps!

how to deal with NA in neural network prediction result in R

1 Answers