Precision focuses on the proportion of times a model said the result was positive and the result was actually positive — as opposed to being labeled positive falsely. When you think about precision, focus on positivity. It's about the accuracy of positive predictions.

In terms of binary classification model's confusion matrix, precision = TP / (TP + FP). You can learn more about using precision in multiclassification models on Evidently AI.

It's helpful to think about precision when evaluating the positive instances of a model. Think about a model designed to filter spam emails and regular spam emails. A false positive — or a regular email filtered as spam — isn't great. We don't want very important emails to be labeled as junk. So having a good precision score is important in this scenario in particular.

There are other metrics used to evaluate the effectiveness of a model including F1 Score, recall, and accuracy.