Introduction

Introduction

The confusion matrix holds four numbers. That's informative but hard to compare between models. Precision and recall squeeze it into two numbers, each answering a simple question about the positive class:

• Precision: When the model raises an alarm, how often is it right?
• Recall: Of all the real cases, how many does the model catch?

Think of a fishing net:

• Recall is how many of the fish in the lake you caught.
• Precision is how much of what's in your net is actually fish (rather than boots and plastic bags).

A huge net catches nearly every fish (high recall) but also lots of junk (low precision). A tiny, careful net brings up almost only fish (high precision) but misses most of them (low recall).

We'll keep using the spam filter from topic 2.1: