## Competition Rules (Simplified)

1. **Competition Entry**

   * By participating, you agree to follow these rules.
   * Only **one account per participant** is allowed. Submissions from multiple accounts are not permitted.

2. **Team Limits**

   * Teams can have up to **4 members**.

3. **Submission Limits**

   * Each participant or team can submit a maximum of **5 predictions per day**.
   * Up to **1 submission** can be selected as the final submission for judging.

4. **Data Use**

   * The provided dataset can only be used for this competition.
   * External data and tools are allowed if they are **publicly available** and **equally accessible to all participants**.

---

### **Evaluation Criteria**

Solutions will be assessed in two main areas:

---

### ▌**Model Quality (80%)**

**Evaluation Data:**

* Historical data for bus demand prediction.
* A hidden **out-of-time (OOT) dataset** will be used to assess model generalization.

**Evaluation Procedure:**

* Participants train their models on the provided historical data.
* Predictions are submitted for the OOT period.

**Metrics:**

* **sMAPE (Symmetric Mean Absolute Percentage Error)** — the primary metric for evaluating forecast accuracy.

---

### ▌**Code Quality + Presentation Quality (20%)**

Teams must present their methodology and results to the jury:

* Clarity of the forecasting approach and reasoning behind model selection.
* Visualization of predictions vs. actual data.

---

### **Technical Requirements**

* **Programming language:** Python
* **Allowed frameworks:** Any (e.g., PyTorch, TensorFlow, Scikit-learn, CatBoost, Prophet, etc.)
* **External models:** Integration of LLMs or specialized time-series APIs is permitted
* **Deliverables:**

  * Training notebook/code
  * Forecast submission file
  * Short methodological report (slide deck or interface)