📈 Model Performance [Neurosurgery Wiki]

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== 📈 Model Performance ======

Evaluating **model [[performance]]** is crucial to determine how well a machine learning model generalizes to new data. In medical applications (e.g., lesion classification), this ensures clinical usefulness and patient safety.

===== 🧪 Key Performance Metrics =====

^ Metric ^ Description ^ Interpretation ^
| **Accuracy** | Ratio of correct predictions to total cases | Good for balanced datasets |
| **Precision** | TP / (TP + FP) | How many predicted positives are true |
| **Recall (Sensitivity)** | TP / (TP + FN) | Ability to detect true positives |
| **Specificity** | TN / (TN + FP) | Ability to detect true negatives |
| **F1-Score** | Harmonic mean of precision and recall | Balances precision & recall |
| **AUC (Area Under ROC Curve)** | Measures ability to distinguish between classes | Closer to 1 = better |
| **Balanced Accuracy** | Mean of sensitivity and specificity | Useful for imbalanced datasets |
| **Confusion Matrix** | Table showing TP, FP, TN, FN | Full picture of model errors |

===== 🧠 In Multi-Class Problems =====

* Metrics are computed **per class**, then averaged:
  * **Macro average**: Equal weight to each class
  * **Micro average**: Accounts for class imbalance
  * **Weighted average**: Weighted by class frequency

===== 🛠 Python Snippet (Example with scikit-learn) =====

<code python>
from sklearn.metrics import classification_report, confusion_matrix, roc_auc_score

# Assume y_test = true labels, y_pred = predicted labels
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

# For multi-class AUC (with probabilities)
roc_auc_score(y_test, y_proba, multi_class="ovr")
</code>

===== 📊 Clinical Relevance =====

* **High sensitivity**: Essential in detecting critical conditions (e.g., tumors)
* **High specificity**: Important to avoid false positives
* **Balanced accuracy**: Prevents overestimation in imbalanced data (e.g., rare tumors)

===== ⚠️ Best Practices =====

* Always report **multiple metrics**, not just accuracy
* Use **cross-validation** to avoid overfitting
* Consider **confidence intervals** for key metrics