Binning or Feature Selection: Which One Won Us an Insurance Hackathon?

A winning solution for an insurance hackathon was re-evaluated a year after its success to determine the key factors. The original approach involved factoring claim cost prediction into frequency and severity models and using supervised binning on over 400 features. The audit revealed that while binning was helpful, the specific binning method or library used had a negligible impact on accuracy compared to a rival library. Ultimately, the choice of which features were selected and kept proved to be the more decisive factor in the model's performance, contrary to the team's initial belief.