So you’re reading all these stories about machine learning being used to train models for predicting cancer. You may even read about a model with a 90% accuracy rate.
What if I told you that a model with an accuracy rate of 73% could be a better performer (higher Precision and Recall) than one with 90% accuracy.
You can not draw conclusions on headlines about 90% accuracy without understanding the accuracy paradox, and knowing that we can not rely on accuracy as the only metric.
I’ll let Tejumade Afonja explain, because intuition isn’t always your friend.
”Accuracy Paradox for Predictive Analytics states that Predictive Models with a given level of Accuracy may have greater Predictive Power than Models with higher Accuracy.”