Precision, Recall, and F1 Score Explained: From Basics to Advanced

Precision, Recall, and F1 Score are key evaluation metrics used to measure the performance of machine learning classification models. Unlike accuracy, which only shows overall correctness, these metrics provide deeper insights into how well a model identifies positive cases and handles prediction errors.

They are particularly valuable when working with imbalanced datasets, where one class significantly outnumbers another. In such situations, a model can achieve high accuracy while still performing poorly, making precision, recall, and F1 score more reliable indicators of real-world performance.

In this blog, you will learn what precision, recall, and F1 score mean, how to calculate each one, when to use which metric, and how they all work together. 

Transform your career with upGrad’s Data Science Course. Learn from industry experts, work on hands-on projects, and gain the skills top employer’s demand.

What Is Precision, Recall, and F1 Score?

Before jumping into formulas, let us look at the problem these metrics solve.

Imagine a spam email classifier. You test it on 1,000 emails. 950 are normal and 50 are spam. If your model predicts every single email as "not spam," it is 95% accurate. But it catches zero spam emails. That is a useless model.

Accuracy precision recall f1 score exist to expose exactly this kind of failure.

To understand these metrics, you first need to know four terms:

Term	What It Means	Example
True Positive (TP)	Correctly predicted positive	Spam email correctly flagged as spam
True Negative (TN)	Correctly predicted negative	Normal email correctly marked as not spam
False Positive (FP)	Predicted positive but actually negative	Normal email wrongly marked as spam
False Negative (FN)	Predicted negative but actually positive	Spam email wrongly marked as safe

These four values form the foundation of precision recall and f1 score calculations.

Also Read: What is Classification in Machine Learning?

Precision

Precision answers one question: Of all the cases your model predicted as positive, how many were actually positive?

Precision Formula:

Precision = TP / (TP + FP)

If your spam classifier flags 100 emails as spam and 80 of them are actually spam, your precision is 80%.

High precision means fewer false alarms. When false positives are costly, like in legal document review or fraud detection, you want high precision.

Recall

Recall answers a different question: Of all the actual positive cases, how many did your model catch?

Recall Formula:

Recall = TP / (TP + FN)

If there were 50 real spam emails and your model caught 40 of them, your recall is 80%.

High recall means fewer misses. When missing a positive case is dangerous, like in cancer diagnosis or fraud detection, recall becomes more important.

Also Read: How to Perform Cross-Validation in Machine Learning?

The Precision-Recall Trade-off

Here is the tricky part. Precision and recall usually pull against each other.

If you lower your classification threshold to catch more positives (more spam), recall goes up but precision drops. If you raise the threshold to be more selective, precision improves but recall falls.

This is the classic precision-recall trade-off, and it is one of the most important things to understand in model evaluation.

The F1 Score Formula and How to Calculate It

The F1 score is the solution to the precision-recall trade-off problem. It combines both into a single number.

The precision recall f1 score formula uses the harmonic mean of precision and recall. Why harmonic mean? Because it penalizes extreme values. A model with 100% precision and 0% recall would have an arithmetic mean of 50%, which sounds decent but is clearly broken. The harmonic mean gives a much more honest result.

Also Read: What Is Data Science? Courses, Basics, Frameworks & Careers

F1 Score Formula:

F1 = 2 × (Precision × Recall) / (Precision + Recall)

F1 Score Calculation Example

Let us work through a real example.

You have a model that detects fraudulent transactions:

Metric	Value
True Positives (TP)	70
False Positives (FP)	30
False Negatives (FN)	20

Step 1: Calculate Precision

Precision = 70 / (70 + 30) = 70 / 100 = 0.70

Step 2: Calculate Recall

Recall = 70 / (70 + 20) = 70 / 90 = 0.78

Step 3: Calculate F1 Score

F1 = 2 × (0.70 × 0.78) / (0.70 + 0.78) = 2 × 0.546 / 1.48 = 0.74

The F1 score is 0.74, which means the model is reasonably well-balanced but has room to improve.

F1 Score Range

F1 scores range from 0 to 1:

Score	Interpretation
0.90 to 1.00	Excellent
0.80 to 0.89	Good
0.70 to 0.79	Acceptable
Below 0.70	Needs improvement

Also Read: Career in Data Science: Jobs, Salary, and Skills Required

When to Use Precision vs Recall vs F1 Score

This is where most beginners get confused. Each metric serves a different purpose. Choosing the wrong one can lead to bad decisions about your model.

Use Precision When

False positives are expensive. If your model wrongly flags a good customer as fraudulent, they get locked out of their account. That damages trust.

Use precision when:

• Email spam filtering (a false positive loses important emails)
• Legal document classification
• Medical test result systems where unnecessary treatment is harmful

Use Recall When

False negatives are expensive. Missing a real case is worse than having a few false alarms.

Use recall when:

• Cancer or disease detection (missing a positive case can cost a life)
• Security threat detection
• Quality control in manufacturing (missing a defect means a faulty product ships)

Also Read: Real Data Science Case Studies That Drive Results!

Use F1 Score When

You want a balanced view and the cost of false positives and false negatives is roughly equal. The F1 score is especially useful when:

• Your dataset is imbalanced
• You cannot afford to prioritize one error type over another
• You are comparing multiple models and want one number to rank them

Accuracy vs Precision Recall F1 Score

Metric	Best Used When
Accuracy	Dataset is balanced, all errors cost the same
Precision	False positives are very costly
Recall	False negatives are very costly
F1 Score	Balance is needed, dataset is imbalanced

When your data is skewed, accuracy precision recall f1 score analysis together gives a much more complete picture than accuracy alone.

Advanced Concepts: Macro, Micro, and Weighted F1 Scores

Once you move beyond binary classification (positive vs negative), things get more complex. Multi-class problems need a way to average F1 scores across all classes. This is where macro, micro, and weighted F1 come in.

1. Macro F1 Score

Calculates F1 for each class independently and then takes the simple average.

Macro F1 = (F1 class 1 + F1 class 2 + ... + F1 class N) / N

This treats every class equally, regardless of how many examples it has. Good when all classes matter equally.

Also Read: Introduction to Classification Algorithm: Concepts & Various Types

2. Micro F1 Score

Aggregates the total true positives, false positives, and false negatives across all classes, then calculates a single F1 score from those combined counts.

Good when you care about overall performance across all examples, especially with class imbalance.

3. Weighted F1 Score

Calculates F1 for each class but weights it by the number of actual examples in that class. Good when some classes naturally appear more often and you want to reflect that in your evaluation.

Which One Should You Use?

Scenario	Recommended
All classes equally important	Macro F1
Class imbalance, care about total correct	Micro F1
Some classes more common, weight matters	Weighted F1

The F-Beta Score

The F1 score assumes precision and recall are equally important. But that is not always true. The F-beta score lets you control this balance.

F-beta = (1 + beta²) × (Precision × Recall) / ((beta² × Precision) + Recall)

When beta is greater than 1, recall matters more. When beta is less than 1, precision matters more. F1 is simply F-beta where beta equals 1.

Also Read: Binary Logistic Regression: Concepts, Implementation, and Applications

Precision, Recall, and F1 Score in Python

Let us see how to calculate precision recall and f1 score using scikit-learn, which is the most common library for this in Python.

from sklearn.metrics import precision_score, recall_score, f1_score, classification_report

# Actual labels
y_true = [0, 1, 1, 0, 1, 1, 0, 0, 1, 0]

# Predicted labels
y_pred = [0, 1, 0, 0, 1, 1, 1, 0, 1, 0]

# Calculate individual metrics
precision = precision_score(y_true, y_pred)
recall = recall_score(y_true, y_pred)
f1 = f1_score(y_true, y_pred)

print(f"Precision: {precision:.2f}")
print(f"Recall:    {recall:.2f}")
print(f"F1 Score:  {f1:.2f}")

# Full report for multi-class
print(classification_report(y_true, y_pred))

The classification_report function is especially useful. It gives you precision, recall, and F1 score for every class in your dataset, along with support (number of examples per class).

For multi-class averaging:

# Macro averaging
f1_macro = f1_score(y_true, y_pred, average='macro')

# Weighted averaging
f1_weighted = f1_score(y_true, y_pred, average='weighted')

Conclusion

Precision, recall, and F1 score are not just numbers to fill in a report. They tell you what kind of mistakes your model is making and whether those mistakes are acceptable for your use case.

To quickly recap:

• Precision measures how clean your positive predictions are
• Recall measures how complete your positive coverage is
• F1 score balances both into a single, honest metric
• Accuracy alone is often misleading, especially with imbalanced data

Once you understand the precision recall f1 score formula, you will never look at a model's accuracy number the same way again. These metrics give you the real story.

Want personalized guidance in Data Science and upskilling? Speak with an expert for a free 1:1 counselling session today.