8.4.3. Recall (Sensitivity)

Recall (Sensitivity)

• Measures how many actual positives the model correctly predicted.

$\text{Recall} = \frac{\text{True Positives}}{\text{True Positives} + \text{False Negatives}}$

Recall, also known as sensitivity, is a performance metric that evaluates the effectiveness of a classification model by calculating how well it identifies actual positive cases from a dataset. It is defined by the formula: Recall equals the number of true positive predictions divided by the sum of true positives and false negatives. True positives are instances correctly predicted as positive, while false negatives are actual positives incorrectly predicted as negatives. Therefore, recall indicates the ability of the model to capture all relevant cases.

Imagine you are a doctor trying to diagnose a disease. True positives are the patients correctly diagnosed with the disease, while false negatives are those who actually have the disease but were not diagnosed. High recall means you are good at recognizing most patients who have the disease, which is crucial for treatment.

Recall is particularly important in scenarios where missing a positive case has significant consequences. For example, in medical diagnoses, failing to identify a sick patient (false negative) can lead to severe outcomes.

In various applications, recall plays a crucial role, especially in fields like healthcare, fraud detection, and spam detection. A high recall indicates that the model is responsive to identifying true positive cases, minimizing the chances of overlooking vital instances. This is important in situations where false negatives can lead to negative consequences. A medical test for a disease needs high recall so that most patients with the condition are identified and treated, avoiding serious health implications.

Think of a fire alarm system. If the alarm fails to go off when there's a fire (false negative), it can lead to disastrous and life-threatening consequences. Therefore, having a system that reliably detects fire (high recall) is essential for safety.

Recall is often compared to metrics like precision. While recall focuses on capturing all actual positives, precision measures the accuracy of positive predictions, highlighting different aspects of model performance.

While recall focuses on the proportion of true positives out of the total actual positives, precision looks at how many of the predicted positive cases are correct. This means that a model can have high recall while still having low precision if it makes a lot of incorrect positive predictions. Understanding the trade-off between recall and precision is vital; sometimes, prioritizing one over the other is necessary based on specific requirements of the application.

Consider a fishing net. If the net has large holes, it will catch big fish (high recall) but allow small ones to escape (low precision). In contrast, a fine mesh might catch only the small fish, leading to low recall but high precision. Depending on whether you want to catch all fish (high recall) or only the big ones (high precision), you'll choose the type of net accordingly.

--