Trees & Ensembles
A loan officer never runs one formula; they ask income > 50000?, then a follow-up, then another, each answer narrowing toward approve or deny. A hundred officers who each ask different questions and pool their calls beat any single one.
- ▸A prediction can be a chain of yes/no questions you trace by finger, one answer at a time.
- ▸Change a few training examples and a model can rebuild its whole chain of questions, flipping verdicts on unchanged cases.
- ▸A crowd of mediocre models that disagree beats one careful model: their scattered mistakes cancel while the shared answer survives.
An application for an $18,000 car loan lands on the desk, and instead of one calculation it faces a chain of plain questions. Is the income above $50,000? Yes, go right. Is there more than $30,000 of existing debt? No, go left. Two years on the job? Yes. Each answer picks a branch, and following those branches leads to an endpoint that reads approve or deny. That endpoint is a leaf, and the whole structure of questions and branches is a decision tree.
Nothing about the applicant ever enters a formula. The details matter only through which path they send the application down, and the route from the first question to the leaf it lands on is the entire prediction. Because of that, the whole verdict can hang on a single branch: nudge one number across a cut-off and the path can divert to a different leaf and a different answer.