📊 What is Model Evaluation?
Model Evaluation is the process of measuring how well a machine learning model performs on unseen data. It helps determine whether your model is accurate, reliable, and generalizes well to new inputs.
📌 Without evaluation, you can’t trust a model — even if it performs well during training.
🎯 Why Model Evaluation is Important
- 🧪 Detects overfitting or underfitting
- 📏 Measures accuracy and performance
- 🏆 Helps select the best model among many
- ⚙️ Supports hyperparameter tuning
📁 Types of Evaluation Based on Task
| Task | Common Metrics |
|---|---|
| Classification | Accuracy, Precision, Recall, F1 Score, AUC-ROC |
| Regression | Mean Absolute Error (MAE), Mean Squared Error (MSE), R² |
| Clustering | Silhouette Score, Davies–Bouldin Index, Inertia |
✅ Key Metrics for Classification Models
| Metric | Description |
|---|---|
| Accuracy | % of correct predictions (good for balanced classes) |
| Precision | True Positives / (True Positives + False Positives) – focuses on relevance |
| Recall | True Positives / (True Positives + False Negatives) – focuses on completeness |
| F1 Score | Harmonic mean of Precision and Recall (good for imbalanced data) |
| AUC-ROC | Area under the curve – measures classification confidence |
📏 Key Metrics for Regression Models
| Metric | Description |
|---|---|
| MAE | Average of absolute differences between prediction and true values |
| MSE / RMSE | Penalizes larger errors more heavily |
| R² (R-squared) | Measures how well the variance in target is explained by features |
🔁 Train-Test Split vs Cross-Validation
| Method | Use Case |
|---|---|
| Train/Test Split | Quick checks, typically 80/20 or 70/30 splits |
| K-Fold Cross-Validation | More reliable; splits data into K parts and rotates evaluation |
| Stratified K-Fold | Keeps class balance in each fold (for classification) |