Elastic Net vs. Ridge/Lasso

I thought Bisara’s work on Ridge / Lasso was really good. After reading that, I had a hypothesis that using elastic net regularization would do even better. However, when I tried it, I got worse results. Why? My kernel is here: https://www.kaggle.com/hueykwik/titanic/elastic-net-implementation Some notes: – I used kappa as the metric because that’s what the original author was using. To do that, I had to use multiClassSummary as my summaryFunction. – I did try metric = ROC and summaryFunction = twoClassSummary, but I remember getting a worse score (~75%) when I did the submission. My guess is that I’m simply doing something wrong, but if there’s a property of elastic net that would explain my results, that would be cool to know, too.

Link to Full Article: Elastic Net vs. Ridge/Lasso