# Interpreting the outcomes of elastic net regression for binary classification

I am using elastic net for binary classification, where the outcomes are either 1 or 0 (they could also be 1 or -1, but I am not sure which one is better).

The output of the net is real valued though, and I am not sure how to interpret them. any ideas?

It seems like a sensible thing to do is to say anything larger than 0.5 is classified as 1, and 0 otherwise. But I certainly cannot interpret this as a probability correct? In fact many of the values are less than 0.

• Note that for binary classification you can use elastic-net with logistic regression (e.g. see here), which would be more interpretable. – GeoMatt22 May 1 '17 at 1:06
• I agree with GeoMatt above. It is possible that you are doing things correctly, and are producing predictions on the linear scale, i.e. without applying the logistic link function. Also, it is generally not a good practice to threshold the predicted probabilities at 0.5. Either evaluate your model as it is, where the predictions are probability, or determine the relative costs of false positives and false negatives, and threshold your classification to minimize some cost function. – Matthew Drury May 1 '17 at 2:54