---

## 📊 Evaluation

Submissions will be evaluated based on the **F1 Score**, a metric that balances **precision** and **recall**, especially useful when dealing with **imbalanced datasets** like customer churn.

### 🎯 Scoring Metric

We will use the **macro F1 Score**, which calculates the F1 score for each class (churned / not churned) independently and takes the average, treating both classes equally regardless of their support.

$$
\text{F1}_{macro} = \frac{1}{2}(\text{F1}_{class=0} + \text{F1}_{class=1})
$$

This ensures your model performs well not just on the majority class (non-churned customers), but also on the minority class (churned customers).

### 📝 Submission Format

Your final predictions should be submitted in a `.csv` file with the following format:

```
is,loan_defaulted
768805383,0
818770008,1
819686202,0
...
```

* `id`: The unique ID for each customer from the test set.
* `loan_defaulted`: Your model’s predicted class (0 = not churned, 1 = churned).

Make sure your submission contains predictions for **every row** in the test set and follows the exact format shown above.

---