Hands-On Recommendation Systems with Python

Binary classification metrics

Sometimes, accuracy does not give us a good estimate of the performance of a model.

For instance, consider a binary class dataset where 99% of the data belongs to one class and only 1% of the data belongs to the other class. Now, if a classifier were to always predict the majority class for every data point, it would have 99% accuracy. But that wouldn't mean that the classifier is performing well.

For such cases, we make use of other metrics. To understand them, we first need to define a few terms:

• True positive (TP): True positive refers to all cases where the actual and the predicted classes are both positive
• True negative (TN): True negative refers to all cases where the actual and the predicted classes are ...