I am an undergrad student and I'm super new to R! I have a data set that I have split into a training and test set. I obtained a multiple regression model from my training set, and now I want to use it to predict my test data. My dependent variable is Plant Species Richness (PSR), and my original data set had 4 independent variables (Area, AdjacentWetlands, Roads, and Forest) but my model is only using Area and Forest: LM<-lm(PSR~Area+Forest, data=Wetlands). How do I use this model to predict PSR in my test set? And then how do I assess whether it is a good prediction or not?
$\begingroup$
$\endgroup$
2
-
1$\begingroup$ Split-sample validation takes an amazingly large sample size to work in the sense of providing almost the same answer if you were to repeat the single random split and re-do all model building and all validation calculations. What is your total sample size? $\endgroup$– Frank HarrellCommented Sep 22, 2015 at 19:07
-
1$\begingroup$ If you're interested in learning about machine learning and R, I'd highly recommend checking out An Introduction to Statistical Learning, which is a great book that is freely available as a pdf. It has many examples in R, including cross validation (Chapter 5) $\endgroup$– TchotchkeCommented Sep 22, 2015 at 19:27
Add a comment
|
3 Answers
$\begingroup$
$\endgroup$
2
Firstly, get your model:
LM <- lm(PSR ~ Area+Forests, data = Wetlands)
Make sure all data values are correct.
The function predict() does the calculation:
pred <- pred(your_model, your_data_test)
Your issue seems that your_data_test have more variables than your model, right?
So you can slice your_data_test and put into a new_data_test by using
new_data_test <- data.frame(your_data_test$variable1,your_data_test$variable2)
and then
pred <- pred(yourmodel, new_data_test)
I suppose should be work for you.
-
$\begingroup$ I tried your solution on a similar dataset, but this gives a warning: " 'newdata' had 15600 rows but variables found have 1164007 rows " Also the created predict vector is not of 15600 length as it should be, but of 1164007 length. $\endgroup$– ApurvCommented May 25, 2017 at 13:15
-
$\begingroup$ Please check your data and try to catch any inconsistency in the values. $\endgroup$ Commented Nov 28, 2019 at 17:25
$\begingroup$
$\endgroup$
Actually, you don't need to create new_data_set. Instead, simply use your_data_test to get pred since your_model (LM) restricts to its entries (2 variables as mentioned above).
$\begingroup$
$\endgroup$
1
You can use print(summary(LM)) to access the accuracy of your model. You'll also get to know of other vitals like r^2 value, p-value of each predictor which helps you decide on a model.
-
$\begingroup$ this is true and generally helpful, but the OP asks "How do I use this model to predict PSR in my test set? And then how do I assess whether it is a good prediction or not?" $\endgroup$– JimCommented Feb 20, 2018 at 9:33