US Accidents for Domain Generalization

A preprocessed version of Kaggle's "US Accidents (2016 - 2023)" dataset. The dataset is split by cities and is meant to test domain generalization. A cleaned and processed version of the "US Accidents dataset (2016–2023)" on Kaggle, where the task is to predict accident severity via estimated traffic delay duration. Features include weather, road, and time-related conditions. Domains correspond to different U.S. states: CA, TX, FL, OR, and MN. The dataset is used to evaluate domain generalization.