# FloodHackathon — Data

来源：https://www.kaggle.com/competitions/floodhackathon/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

That’s totally fine 👍 Kaggle allows you to launch a competition **without a dataset at first** — but you’ll need to clearly communicate it to participants. The key is to write a strong placeholder **Dataset Description** so people understand what’s coming.

Here’s a draft you can directly use for your **FloodHackathon Kaggle page**:

---

# 🌊 FloodHackathon Dataset

⚠️ **Note:** This competition does not have a dataset yet.

We are in the process of preparing and curating a dataset that reflects the challenges of **flood prediction, response, and recovery** in Pakistan. The dataset will either be:

1. **Synthetic (simulated)** – generated from historical weather and hydrology patterns to provide a training ground for models.
2. **Open-source based** – built from publicly available rainfall, river flow, and satellite data.
3. **Hybrid** – a combination of synthetic data and real-world references, designed to be both realistic and safe to share.

---

## 📑 Planned Dataset Features

The dataset will likely include the following variables:

| Column              | Description                                                                   |
| ------------------- | ----------------------------------------------------------------------------- |
| `id`                | Unique identifier for each observation                                        |
| `district`          | District or region name                                                       |
| `rainfall_mm`       | Rainfall (mm) during the period                                               |
| `river_flow_cumecs` | River discharge (cubic meters per second)                                     |
| `soil_saturation`   | Soil moisture index (0 = dry, 1 = fully saturated)                            |
| `temp_c`            | Average temperature (°C)                                                      |
| `month`             | Month of observation                                                          |
| `flood`             | Target variable (1 = flood, 0 = no flood) — *available only in training data* |

---

## 🎯 Task

Once released, participants will:

* Train models on **Train.csv** (with labels).
* Predict flood occurrence on **Test.csv** (without labels).
* Submit predictions as **submission.csv** in the required format.

**Evaluation Metric:** ROC-AUC (Area Under the Receiver Operating Characteristic Curve).

---

## 🚀 Why No Dataset Yet?

* We want participants to **focus first on the challenge** and ideas.
* The dataset is being finalized to ensure **quality, realism, and fairness**.
* A notification will be sent to all registered participants as soon as it is published.

---

👉 Until then, participants can:

* Brainstorm possible **features and data sources**.
* Explore **open datasets** on rainfall, river flows, or satellite imagery (e.g., from NASA, NOAA, or Pakistan Meteorological Department).
* Get ready to test their solutions as soon as the dataset is live.

---

This way, your competition page looks **professional and intentional**, even without a dataset right now.

Do you want me to also create a **temporary synthetic mini-dataset (like 50 rows CSV)** that you can upload so the competition pipeline (submissions/evaluation) is already testable?