Never Miss Important Cases
Never Miss Important Cases
In the previous module, we learned that Precision measures how trustworthy the AI's positive predictions are. But consider a different situation. Imagine an AI that helps doctors detect cancer from medical scans. Suppose there are 100 patients who actually have cancer. The AI correctly identifies 92 of them, but fails to detect the remaining 8 patients. From a medical perspective, those eight missed cases could have serious consequences. This brings us to another important evaluation metric called Recall. Recall answers a different question from Precision. Instead of asking, "When the AI says yes, is it correct? " Recall asks, "Out of all the cases that actually deserved a yes, how many did the AI successfully find? " To understand Recall, we again refer to the Confusion Matrix. A True Positive is a case where the AI correctly predicts Yes. A False Negative is a case where the AI predicts No, even though the correct answer was Yes. Recall focuses only on these two outcomes. It measures the model's ability to identify all the positive cases. In our example, the AI found 92 out of 100 cancer patients. Its Recall is therefore 92%. A high Recall means the model misses very few important cases. This is especially valuable in situations where failing to identify a positive case is far more dangerous than raising an unnecessary alert. Consider some examples. A medical diagnosis system should identify as many patients with a disease as possible. A wildfire detection system should detect every possible fire before it spreads. A cybersecurity system should identify every potential attack before it reaches the network. In each of these applications, missing a real case can have severe consequences. That is why Recall becomes the preferred evaluation metric. So when should you use Recall? Use Recall when the cost of a false negative is high—when incorrectly saying "No" is more dangerous than incorrectly saying "Yes. " Recall does not tell us whether every positive prediction is correct. Instead, it tells us how effectively the model discovers all the positive cases that truly exist. Precision and Recall are often discussed together because they measure two different aspects of model performance. One measures trust. The other measures coverage. Understanding how to balance these two metrics is one of the most important skills in machine learning. That balance is exactly what we'll explore in the next module.