**Heart Disease Risk Prediction**

---

**Context**
Heart disease is the leading cause of death worldwide, and early detection of risk factors is crucial to saving lives. This synthetic (yet realistic) dataset simulates medical and lifestyle characteristics of patients, with the goal of predicting whether an individual has a **high risk (1)** or **low risk (0)** of developing heart disease within the next 5 years.

The dataset was programmatically generated in Python, where variable distributions depend on the *target* to replicate causal relationships observed in real medical studies. 

---

**Objective:** Use the data in `train.csv` to train a model capable of generating predictions for the data in `test.csv`. These predictions will be evaluated using the **F1-score**.

**Challenges for Competitors**

* **Complex interactions:** Risk does not depend on a single variable, but on combinations (e.g., smoker + gender + age).
* **Ordinal categorical variables:** Require proper encoding (e.g., `stress_level`).
* **Noise:** The distributions overlap, preventing perfect linear separation.

---

**Evaluation**

* *Main metric:* **F1 Score**

The **F1 Score** measures a model’s performance by balancing two key aspects:

1. **Precision:** The proportion of correctly predicted positive cases (avoiding false alarms).
   `Precision = True Positives / (True Positives + False Positives)`

2. **Recall (Sensitivity):** The proportion of actual positive cases correctly identified (avoiding missed risks).
   `Recall = True Positives / (True Positives + False Negatives)`

**F1 Score formula:**
F1 = 2 × (Precision × Recall) / (Precision + Recall)

**Why is it important?**

* **Contexts with class imbalance** (e.g., 60% “low risk”, 40% “high risk” in your dataset).
* **Problems where both types of errors matter:**

  * *False positives:* Classifying healthy people as at-risk (unnecessary stress/treatment).
  * *False negatives:* Failing to detect those truly at risk (serious consequences).

**In this competition**, a good F1 Score means your model:
✅ Detects most high-risk cases (high Recall).
✅ Minimizes wrong diagnoses (high Precision).

This metric seeks a practical balance between prediction quality and coverage.

---

**Files**

1. `train.csv`: Training dataset (3,500 records).
2. `test.csv`: Dataset for generating predictions (1,500 records).
3. `sample_submission.csv`: Required submission format.


* The submission file must contain **exactly** the following columns: `id` and `heart_disease_risk`.
* Example: (see the sample_submission.csv as well)

  ```
  id,heart_disease_risk
  ID_35001, 1
  ID_35002, 0
  ...
  ```

---

# **Limits:**

* Maximum of **10** submissions per day.
* **Deadline:** December 7, 2025
---

**Acknowledgements**
*Dataset generated by Emerson León – Universidad Antonio Nariño for educational purposes. Inspired by findings from the Framingham Heart Study.*