F-Measure

🧠 What is the F-Measure?

🔹 Definition

The F-measure (F-score) is a metric that combines:

• Precision
• Recall

into a single value.

It is especially useful when there is a tradeoff between precision and recall.

---

📊 Why Do We Need F-Measure?

Accuracy can be misleading in many cases (e.g., imbalanced data).

Instead, we care about:

• How many predicted positives are correct → Precision
• How many actual positives are detected → Recall

👉 F-measure balances both.

---

📐 F1 Score (Most Common Form)

$F1 = 2 \cdot \frac{\text{Precision} \cdot \text{Recall}}{\text{Precision} + \text{Recall}}$

---

🔍 Interpretation

• Harmonic mean of precision and recall 
• Gives equal importance to both

---

🎯 Key Property

• High only when both precision and recall are high
• Penalizes extreme imbalance between them

---

⚖️Why Harmonic Mean?

Instead of arithmetic mean:

• Arithmetic mean ignores imbalance
• Harmonic mean is sensitive to low values

👉 If either precision or recall is low → F1 drops significantly

---

📊  Example

Precision	Recall	F1 Score
1.0	0.0	0
0.9	0.9	0.9
0.9	0.1	Low (~0.18)

👉 Shows balance requirement

---

🔁 General Form: Fβ Score

---

📐 Formula

$F_\beta = (1 + \beta^2) \cdot \frac{\text{Precision} \cdot \text{Recall}}{\beta^2 \cdot \text{Precision} + \text{Recall}}$

🔍 Role of β (Beta)

• Controls importance of recall vs precision

---

📊 Interpretation

β Value	Emphasis
β = 1	Equal (F1)
β > 1	Recall more important
β < 1	Precision more important

---

🎯 Common Variants

---

🔹 F1 Score (β = 1)

• Equal weight to precision and recall

---

🔹 F2 Score (β = 2)

• Recall emphasized
• Used when missing positives is costly

---

🔹 F0.5 Score (β = 0.5)

• Precision emphasized
• Used when false positives are costly

---

🧠 Intuitive Understanding

---

Precision Focus:

“When I predict positive, I want to be correct”

Recall Focus:

“I want to find all positives”

---

F-Measure:

“Balance both objectives”

---

📊When to Use F-Measure

---

✅ Suitable When:

• Class imbalance exists
• Both FP and FN matter
• Need a single metric

---

❌ Not Ideal When:

• True negatives are important
• Probabilities are required

---

🔁 Multi-Class F-Measures

---

🔹 Macro F1

• Compute F1 per class
• Average equally

👉 Treats all classes equally

---

🔹 Micro F1

• Aggregate TP, FP, FN globally
• Then compute F1

👉 Dominated by majority class

---

🔹 Weighted F1

• Weighted by class frequency

👉 Balances class importance

---

📊 Comparison of Averaging

Method	Focus
Macro	Minority classes
Micro	Overall performance
Weighted	Balanced

---

🧪  Practical Example

Suppose:

• High precision, low recall → model misses many positives
• High recall, low precision → many false alarms

👉 F1 gives a balanced score

---

⚠️ Limitations of F-Measure

---

🔹 Ignores True Negatives

• Does not consider TN
• Not suitable for some applications

---

🔹 Equal Weight Assumption (F1)

• May not match real-world costs

---

🔹 Threshold Dependent

• Depends on classification threshold

---

📚 Significance in Machine Learning

F-measure is widely used in:

• Information retrieval
• NLP tasks (e.g., named entity recognition)
• Medical diagnosis
• Imbalanced classification

---

🎯Key Takeaways

• F-measure combines precision and recall
• F1 is most commonly used
• Fβ allows flexible weighting
• Essential for imbalanced datasets
• Must be chosen based on application needs

🧮 Example 1: Basic F1 Score Calculation

📊 Given Confusion Matrix

	Predicted +	Predicted -
Actual +	TP = 40	FN = 10
Actual -	FP = 20	TN = 30

---

🔹 Step 1: Compute Precision

$$\text{Precision} = \frac{TP}{TP + FP} = \frac{40}{40 + 20} = \frac{40}{60} = 0.67$$

---

🔹 Step 2: Compute Recall

$$\text{Recall} = \frac{TP}{TP + FN} = \frac{40}{40 + 10} = \frac{40}{50} = 0.80$$

---

🔹 Step 3: Compute F1 Score

$F1 = 2 \cdot \frac{0.67 \cdot 0.80}{0.67 + 0.80}$

$$F1 = 2 \cdot \frac{0.536}{1.47} \approx 0.73$$

---

✅ Final Answer:

• Precision = 0.67
• Recall = 0.80
• F1 Score ≈ 0.73

---

🧮 Example 2: Imbalanced Scenario

📊 Given

	Predicted +	Predicted -
Actual +	TP = 10	FN = 40
Actual -	FP = 5	TN = 945

---

🔹 Step 1: Precision

$$\frac{10}{10 + 5} = \frac{10}{15} = 0.67$$

---

🔹 Step 2: Recall

$$\frac{10}{10 + 40} = \frac{10}{50} = 0.20$$

---

🔹 Step 3: F1 Score

$$F1 = 2 \cdot \frac{0.67 \cdot 0.20}{0.67 + 0.20}$$ $$= 2 \cdot \frac{0.134}{0.87} \approx 0.31$$

---

✅ Insight

• Precision looks good (0.67)
• But recall is poor (0.20)
  👉 F1 drops to 0.31, revealing poor performance

---

🧮 Example 3: Fβ Score (Recall Emphasis)

---

📊 Given:

• Precision = 0.6
• Recall = 0.9
• β = 2 (Recall more important)

---

🔹 Step 1: Apply Formula

$F_2=(1+2^2)\cdot \frac{0.6\cdot 0.9}{2^2\cdot 0.6+\mathrm{0.9}}$

---

🔹 Step 2: Solve

$$F_2 = 5 \cdot \frac{0.54}{2.4 + 0.9} = 5 \cdot \frac{0.54}{3.3} = 5 \cdot 0.164 = 0.82$$

---

✅ Interpretation

• High recall → boosted score
• F2 = 0.82 (higher than F1 would be)

---

🧮 Example 4: F0.5 Score (Precision Emphasis)

---

📊 Given:

• Precision = 0.9
• Recall = 0.5
• β = 0.5 (Precision more important)

---

🔹 Step 1: Formula

$$F_{0.5} = (1 + 0.5^2) \cdot \frac{0.9 \cdot 0.5}{0.5^2 \cdot 0.9 + 0.5}$$

---

🔹 Step 2: Solve

$$= 1.25 \cdot \frac{0.45}{0.225 + 0.5} = 1.25 \cdot \frac{0.45}{0.725} \approx 1.25 \cdot 0.62 = 0.78$$

---

✅ Interpretation

• Precision is high → score is boosted
• F0.5 = 0.78

---

🧮 Example 5: Comparing Two Models

---

📊 Model A:

• Precision = 0.8
• Recall = 0.6

📊 Model B:

• Precision = 0.6
• Recall = 0.9

---

🔹 Compute F1

Model A:

$$F1 = 2 \cdot \frac{0.8 \cdot 0.6}{0.8 + 0.6} = 2 \cdot \frac{0.48}{1.4} = 0.69$$

---

Model B:

$$F1 = 2 \cdot \frac{0.6 \cdot 0.9}{0.6 + 0.9} = 2 \cdot \frac{0.54}{1.5} = 0.72$$

---

✅ Conclusion

• Model B has higher F1
• Better balance between precision and recall

---

🎯 Key Teaching Insights

• F1 penalizes imbalance between precision and recall
• Fβ allows task-specific weighting
• Useful in imbalanced datasets
• Always interpret alongside precision & recall