Emil O W Kirkegaard

First six questions selected by model: 1. I am the life of the party 9. I am relaxed most of the time 5. I have a vivid imagination 18. I make a mess of things 17. I am not really interested in others 4. I have frequent mood swings

As one would expect when enforcing sparcity, the model selects a question from each of the big5 before fleshing out important factors (starting with Neuroticism). The first five order is ENOCA

The dataset is 64% female so it almost always best to predict female. Therefore I weighted the male samples so that from the model's perspective a random guess is as good as guessing the mode. (When n is close to infinity equivalent to duplicating some male samples)

Notice how q1 "I am the life of the party" (extroversion) and q9 "I am relaxed most of the time" (neuroticism) are doing a lot of the work. The next big jump in accuracy comes when adding q17 "I am not really interested in others" (aggreeableness)

With all features, the largest coefs from logistic regression are: q17 I am not really interested in others q4 I have frequent mood swings q7 I am not interested in other people’s problems Overall I was surprised at only 65% accuracy with 20 questions.