# RoadSense: Predict the Severity of Road Accidents — Data

来源：https://www.kaggle.com/competitions/etiq-roadsense/data

采集时间：2026-10-19T16:27:17Z

The dataset for this competition is based on real road accident data and has been intentionally altered to present a more challenging data-cleaning and feature-engineering task. The data is split across four files, which you will need to merge to create your final training dataset.

## Files

You are provided with four training data files, and four test data files with train or test as their respective suffix. You will need to predict the Gravity for a test set that mirrors the structure of these files.

* ```accidents_train, accidents_test```: The primary file, containing one row for each accident.
 * ```AccidentId```: A unique identifier for each accident.
 * ```Gravity```: The target variable. Your goal is to predict this value ("Lethal" or "NonLethal").
 * ```Date, Hour```: Information about when the accident occurred.
 * ```Light, Weather, CollisionType```: Details about the conditions and nature of the accident.
 * ... and other accident-specific information.
* ```places_train, places_test```: Information about the location and road conditions for each accident.
 * ```AccidentId```: Foreign key to link with the accidents file.
 * ```RoadType, Circulation, SurfaceCondition```: Details about the road where the accident happened.
 * ... and other location-based features.
* ```users_train, users_test```: Information about the individuals involved in each accident. Note that one accident can involve multiple users.
 * ```AccidentId```: Foreign key to link with the accidents file.
 * ```VehicleId```: Foreign key to link with the vehicles file
 * ```Category, Gender, BirthYear```: Demographic and user-type information.
 * ```SafetyDevice, SafetyDeviceUsed```: Information about safety measures.
 * ... and other user-specific data.
* ```vehicles_train, vehicles_test```: Information about the vehicles involved in each accident.
 * ```AccidentId```: Foreign key to link with the accidents file.
 * ```VehicleId```: Primary key to link with the vehicles file.
 * ```Category```: Type of vehicle.
 * ```FixedObstacle, MobileObstacle```: Information about what the vehicle collided with, if anything.
 * ... and other vehicle-specific attributes.

## Data "Features" and Challenges

This dataset is designed to be tricky! Here are some of the challenges you can expect to encounter:

* **Inconsistent Formatting**: IDs, dates, and categorical values may not be uniform across the files.
* **Missing Values**: Data may be missing, sometimes denoted by NULL, NA, -, or just blank cells.
* **Noisy and Redundant Data**: You'll find messy text fields and potentially overlapping information across different files.
* **Complex Merging**: You will need to carefully join all four files to create a comprehensive dataset for modeling.