CS5720 - Week 3
Slide 57 of 60

Model Evaluation Metrics

Why Metrics Matter

Model evaluation metrics are quantitative measures that help us understand how well our neural network performs on unseen data.
Beyond Loss Functions:

• Loss guides training
• Metrics evaluate performance
• Different metrics reveal different strengths/weaknesses
• Choose metrics that match your business goals
⚠️ Common Pitfall
A model with low loss doesn't always mean good performance! Always evaluate with appropriate metrics on test data.

Key Evaluation Metrics

  • 🎯
    Accuracy
    Percentage of correct predictions overall
  • 🔍
    Precision
    Of all positive predictions, how many were correct?
  • 📊
    Recall (Sensitivity)
    Of all actual positives, how many did we find?
  • ⚖️
    F1 Score
    Harmonic mean of precision and recall
  • 📈
    AUC-ROC
    Area under the ROC curve - threshold independent

Metrics Comparison - Click to Explore

Balanced Dataset
95%
Accuracy
Imbalanced Dataset
99%
Misleading Accuracy!
Medical Diagnosis
98%
High Recall Needed
Spam Detection
97%
High Precision Needed
Remember: The best metric depends on your specific problem and the cost of different types of errors!
Prepared by Dr. Gorkem Kar