The Plexus Recall Gauge
Recall, also known as Sensitivity or True Positive Rate (TPR), answers the question: "Of all the items that were actually positive, what proportion did the classifier correctly identify?" It measures the completeness or comprehensiveness of the classifier in finding all positive instances. A high recall score indicates that the classifier has a low rate of False Negatives (FN).
Why is Recall Important?
Focusing on recall is critical in scenarios where the cost of a False Negative is high. A False Negative occurs when the model incorrectly predicts a positive instance as negative. Examples include:
- Medical Diagnosis: Failing to detect a serious disease in a patient who actually has it. This could delay treatment and have severe health consequences.
- Fraud Detection: Missing a fraudulent transaction, leading to financial loss.
- Safety Systems: An autonomous vehicle failing to detect an obstacle, or a security system failing to detect an intruder.
In these cases, high recall is paramount to ensure as few positive instances as possible are missed, even if it means accepting a higher number of False Positives (lower precision).
How the Plexus Recall Gauge Works
The Recall Gauge in Plexus displays the calculated recall score, ranging from 0% to 100%. The formula is:
Recall = True Positives / (True Positives + False Negatives)
The visual segments on the Recall Gauge generally represent standard performance benchmarks. A recall score of 90% means the classifier successfully identified 90% of all actual positive instances. Like precision, the direct interpretation is straightforward, and the segments visually categorize this performance.
Example: Recall Gauge
A recall of 75% indicates that the classifier found 75% of all actual positive instances.
Recall in Action: Example Scenarios
Let's examine recall using our email filter context, where "Prohibited" is the positive class we aim to detect.
The 'Always Safe' Email Filter (Low Recall Example for 'Prohibited')
Strategy: Label ALL emails as 'Safe'. Actual Data: 3% Prohibited, 97% Safe.
No context for interpretation
Poor agreement
Key Insight:
Recall for 'Prohibited' class: 0.0%. Precision for 'Prohibited': N/A (or 0%). This filter misses ALL prohibited emails, resulting in 0% recall for that critical class, despite its high overall accuracy.
The "Always Safe" filter has 0% recall for the "Prohibited" class. This means it fails to identify any of the prohibited emails. While it achieves 97% accuracy by correctly labeling safe emails, it is useless for its primary task of catching prohibited content due to its catastrophic failure in recall for that class.
Contrast: The Goal of High Recall
In critical medical screening (where "Disease Present" is positive), the goal is very high recall. You want to identify as many true cases as possible, even if it means some healthy individuals are flagged for further testing (False Positives, leading to lower precision for the "Disease Present" class).
Recall and Precision: The Trade-off
Recall and Precision often exhibit an inverse relationship. Increasing recall (e.g., by making a classifier more sensitive to positive cases) can sometimes lead to more False Positives, thereby reducing precision.
The F1-score is a common metric that combines precision and recall into a single number (the harmonic mean), providing a balanced measure. Choosing whether to prioritize recall, precision, or a balance depends heavily on the specific application and the consequences of different types of errors.
Key Takeaways for Recall
- Recall (Sensitivity) measures the ability to find all actual positive instances: TP / (TP + FN).
- High recall means a low False Negative rate.
- Crucial when the cost of False Negatives is high.
- The Plexus Recall Gauge displays this score from 0-100%.
- Often considered in conjunction with Precision; the F1-score balances both.