# DustStorm Prediction Challenge — Data

来源：https://www.kaggle.com/competitions/dust-storm-prediction-challenge/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

**Data Preparation:**
The data is provided in CSV format, containing meteorological data and dust storm labels. The training, validation, and test sets will be provided in separate files.

**Feature Engineering:**
Participants can perform feature engineering on the provided meteorological data to extract features that are helpful for dust storm prediction.

**This competition provides three main datasets:**
- Training Set: Contains meteorological data and dust storm occurrence labels. Used for model training.
Time Range: November 15, 2025, 09:00 to November 18, 2025, 09:00
Data Format: CSV file, each row contains a meteorological record and its corresponding dust storm label.

- Validation Set: Contains meteorological data and dust storm occurrence labels.Used for model tuning and validation. Participants can use the validation set to select the best model parameters.
Time Range: November 19, 2025, 09:00 to November 20, 2025, 09:00
Data Format: CSV file, containing meteorological data and dust storm labels.

- Test Set: Used for evaluating model performance. The test set contains only meteorological data without labels. Participants need to generate predictions based on this data and submit them.
Time Range: November 21, 2025, 09:00 to November 22, 2025, 09:00
Data Format: CSV file, containing only meteorological data for participants to make predictions.

**Data Columns:**
The datasets contain the following columns:
timestamp: The timestamp of the data (e.g., 2025-11-15 09:00:00).
WIN_S_Max: Maximum wind speed (unit: m/s).
RHU: Relative humidity (unit: %).
TEM: Temperature (unit: °C).
dust_storm: Dust storm label (1: dust storm occurred, 0: no dust storm occurred) (only present in training and validation sets).

**Data Description:**
Meteorological Data: Provides historical records of wind speed, humidity, temperature, and other variables believed to be closely related to the occurrence of dust storms.
Dust Storm Labels: The dust storm labels are generated based on meteorological data and actual dust storm events, used to train models to predict future dust storms.