Error Metrics for Skewed Classes

Error Metrics for Skewed Classes

These classes are hard to come up with a numerical evaluation for error: Ratio between positive and negative examples is very close to extreme (ie. number of positive examples is much much smaller than negative examples in Cancer classification).
Skewed classes: when we have a lot more examples of one class than the other class
We have to come up with different error evaluation metric: Precision/Recall
Note: rare class associates with y = 1
High precision and high recall: algorithm is doing well

Resources:
https://www.coursera.org/learn/machine-learning/lecture/tKMWX/error-metrics-for-skewed-classes

Comments