multabench/core-text-cls-us-accidents

MUL_TEXT_US_ACCIDENTS

Part of MulTaBench, a benchmark for multimodal tabular learning.

Paper: MulTaBench: Benchmarking Multimodal Tabular Learning with Text and Image · Leaderboard: multabench/leaderboard

• Target: Severity
• Rows: 100001
• Original source: https://www.kaggle.com/datasets/sobhanmoosavi/us-accidents

Load the benchmark splits with multabench.load_split("MUL_TEXT_US_ACCIDENTS", fold=0).

Column	Type
Source	categorical
Start_Time	datetime
End_Time	datetime
Start_Lat	float
Start_Lng	float
End_Lat	float
End_Lng	float
Distance(mi)	float
Description	string
Street	string
City	string
County	string
State	categorical
Zipcode	string
Country	categorical
Timezone	categorical
Airport_Code	string
Weather_Timestamp	datetime
Temperature(F)	float
Wind_Chill(F)	float
Humidity(%)	float
Pressure(in)	float
Visibility(mi)	float
Wind_Direction	categorical
Wind_Speed(mph)	float
Precipitation(in)	float
Weather_Condition	categorical
Amenity	bool
Bump	bool
Crossing	bool
Give_Way	bool
Junction	bool
No_Exit	bool
Railway	bool
Roundabout	bool
Station	bool
Stop	bool
Traffic_Calming	bool
Traffic_Signal	bool
Turning_Loop	bool
Sunrise_Sunset	categorical
Civil_Twilight	categorical
Nautical_Twilight	categorical
Astronomical_Twilight	categorical
Severity	int

Citation

@misc{arazi2026multabenchbenchmarkingmultimodaltabular,
  title={MulTaBench: Benchmarking Multimodal Tabular Learning with Text and Image},
  author={Alan Arazi and Eilam Shapira and Shoham Grunblat and Mor Ventura and Elad Hoffer and Gioia Blayer and David Holzmüller and Lennart Purucker and Gaël Varoquaux and Frank Hutter and Roi Reichart},
  year={2026},
  eprint={2605.10616},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2605.10616},
}