#p-greater-than-n

Question 1: Single choice

mediumAI-checkedLinear Models & Regularization

Ordinary linear regression is fit to 73 potato samples using 200 gene-expression features. On the training data it reaches $R^2 = 1.00$ and MSE $= 0.00$ ; on new samples $R^2$ is only 0.47. What is the explanation?

The model is underfitting the data, so it needs even more gene features1
The genes must be uninformative, since otherwise new-data $R^2$ would also be 12
$R^2$ cannot be computed on new data, so 0.47 is meaningless3
With more features than samples, it can fit the training data exactly: overfitting4

Timed test