Random ForestIntroduction
Meena's decision tree sorted her mangoes well — until one faulty weighing of 15 g changed the whole tree. So she tries a different plan. Instead of trusting one rule, she will ask a whole crowd of helpers and go with the majority.
The trouble is, her helpers are not experts. Each one names the kind of a mango correctly only 60 % of the time — not much better than tossing a coin. Can a crowd of weak guessers be trusted?
Choose the size of the crowd and complete both missions. Each dot is one helper's vote: green when right, red when wrong. The chart below shows how often the majority is right in the long run, for every size of crowd.
Helpers in the crowd:
From 41 helpers who decide on their own, the majority is right more than 90 % of the time; with 101 helpers, 98 % — although each helper alone is right only 60 % of the time. But when everyone copies one friend, a crowd of 101 is no better than that one friend. A crowd only helps when its members make different mistakes. Hold on to that sentence: it is the whole idea of a random forest.