Guy at a Seattle meetup asked me why my model needed 12 features, and I couldn't answer

Robin Hall · 2026-10-20T01:40:03.068913+00:00

Last week I was at a small ML meetup in Seattle and a guy named Dev asked me to walk him through my feature list for a fraud model, and I froze at number 4 because I had no real reason for half of them. I cut the model down to 5 features that night and my false positive rate dropped from 14% to 9% with basically the same accuracy. Why do we keep piling on inputs when a stranger with one question can see the junk we can't?

Topic Shares

Loading…

Guy at a Seattle meetup asked me why my model needed 12 features, and I couldn't answer | AI Innovations - Topic Shares