Glossary

Recall

Learn what Recall is in machine learning, why it matters, and how it ensures AI models capture critical positive instances effectively.

Train YOLO models simply
with Ultralytics HUB

Learn more

Recall is a key performance metric in machine learning and statistics, particularly important in classification and information retrieval tasks. It measures the ability of a model to correctly identify all relevant instances from a dataset. Also known as Sensitivity or the True Positive Rate (TPR), Recall answers the question: "Of all the actual positive instances, how many did the model correctly predict as positive?" High Recall is crucial in scenarios where missing a positive instance (a False Negative) has significant consequences.

Core Concept

Recall focuses on the actual positive cases within a dataset and quantifies how many of them were successfully captured by the model. It is calculated as the ratio of True Positives (TP) – instances correctly identified as positive – to the sum of True Positives and False Negatives (FN) – instances that were actually positive but incorrectly identified as negative. A model with high Recall correctly identifies most of the positive instances. Understanding Recall is essential for evaluating model performance, often visualized using a Confusion Matrix.

Recall Vs. Precision

Recall is often discussed alongside Precision. While Recall measures the proportion of actual positives correctly identified, Precision measures the proportion of predicted positives that were actually correct (TP / (TP + False Positives)). There is often a trade-off between Precision and Recall; optimizing for one can sometimes negatively impact the other.The choice between prioritizing Recall or Precision depends on the specific application:

  • High Recall is preferred when the cost of missing a positive instance (False Negative) is high.
  • High Precision is preferred when the cost of wrongly identifying a negative instance as positive (False Positive) is high.

The F1-score provides a single metric that balances both Precision and Recall.

Real-World Applications

Recall is a critical evaluation metric in many fields:

  • Medical Image Analysis: In tasks like tumor detection using AI, high Recall is vital. Missing a malignant tumor (a False Negative) could delay treatment with severe consequences. Therefore, diagnostic systems are often optimized for high Sensitivity (Recall), even if it means flagging some benign cases for further review (lower Precision). Ensuring comprehensive detection is paramount in AI in Healthcare.
  • Fraud Detection: When building systems to detect fraudulent financial transactions, maximizing Recall is important. Failing to identify a fraudulent transaction (False Negative) can lead to significant financial loss. While minimizing false alarms (False Positives) is also desirable, ensuring that most fraudulent activities are caught often takes precedence.

Relevance in Computer Vision

In computer vision, Recall is essential for evaluating tasks like object detection and image segmentation. For an object detection model like Ultralytics YOLO, Recall indicates how well the model finds all instances of a specific object class within an image. A high Recall means the model rarely misses objects it's supposed to detect. It is commonly used alongside Precision and mean Average Precision (mAP) to provide a comprehensive assessment of detection performance, as detailed in guides on YOLO Performance Metrics. Tools within platforms like Ultralytics HUB help users track these metrics during model training and validation. Understanding Recall helps developers fine-tune models for specific needs, such as ensuring comprehensive detection in security alarm systems. Evaluating performance often involves analyzing metrics derived from a confusion matrix and considering the context of potentially imbalanced datasets.

Read all