====== Multi-Class Classification ======

Multi-class classification is a type of **supervised machine learning** where the model learns to assign an input to **one of three or more possible categories** (classes).

===== 🧠 What It Means =====

* Unlike binary classification (yes/no, positive/negative), multi-class models can handle problems like:
  * Classifying types of brain tumors
  * Identifying disease subtypes
  * Differentiating among several anatomical lesions

Example:
  * Class 0 → Pituitary adenoma
  * Class 1 → Craniopharyngioma
  * Class 2 → Rathke’s cleft cyst
  * Class 3 → Tuberculum sellae meningioma

===== ⚙️ How It Works =====

1. **Input**: Feature vectors (e.g., radiomics from MRI)
2. **Model**: Trained using labeled data with more than two output classes
3. **Prediction**: Model outputs either:
   * A single class label (e.g., 2 = RCC)
   * A probability distribution over all classes

===== 🚀 Algorithms That Support Multi-Class =====

^ Algorithm ^ Strategy Used ^
| XGBoost | Native support via `multi:softmax` or `multi:softprob` |
| Logistic Regression | Multinomial extension |
| Support Vector Machine (SVM) | One-vs-rest or one-vs-one |
| Random Forest | Native multi-class |
| Neural Networks | Softmax activation in output layer |

===== 📏 Evaluation Metrics =====

^ Metric ^ Notes ^
| **Accuracy** | Percentage of correct predictions |
| **Confusion Matrix** | Shows how predictions match actual labels |
| **Precision / Recall / F1-score** | Often calculated per class |
| **AUC (macro/micro average)** | Useful when using probabilistic outputs |

===== 🏥 Medical Use Cases =====

* Differentiating **sellar region lesions** using MRI
* Classifying **skin lesions** or **lung nodules**
* Assigning patients to **risk categories** or **cancer subtypes**
* Diagnosing **neurological conditions** with overlapping symptoms

===== 💡 Tip =====

Multi-class problems may require:
  * **More data per class**
  * **Class balancing** techniques (e.g., SMOTE)
  * Careful **cross-validation** (e.g., stratified k-fold)