{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/9315227506034fc3b8f427c4a463ebc7\" frameborder=\"0\" width=\"1976\" height=\"1482\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1482,"width":1976,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1482,"thumbnail_width":1976,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/9315227506034fc3b8f427c4a463ebc7-22d5ce6891e0983b.gif","duration":936.768,"title":"Predicting Severe US Traffic Accidents, Findings 📊","description":"In this Loom, I summarize our U.S. traffic accident severity project, where given an accident occurs, we predict whether it is severe using weather, road features, and time of day. After cleaning, we used 246,000 records from Kaggle 2016 to 2021, framed as binary classification where severity 3 and 4 are severe. Our best results came from random forest with 72.2% macro F1 and 0.75 severe recall, improving 4.4% F1 versus logistic regression baseline. Key EDA showed severity peaks late at night, and we found temperature strongly relates to severity. I did not request any action from viewers."}