How pbbdydx/vehicle_research works
Project type and purpose
• This is an R-based data-science/ML research pipeline for predicting vehicle collision severity in Pennsylvania.
• It is not a typical client–server or web app but a linear analysis pipeline: raw data → cleaning → feature engineering → model training → stored models.
File structure
Top-level:
• code/
– analysis.R
– data_cleaning.R
– data_pipeline.R
– models.R
• models/ (serialized RDS model outputs)
• README.md, .gitignore, R session files
Observations:
– “code/” contains modular scripts for each stage.
– “models/” holds final artifacts (logistic, random forest, XGBoost, with and without ROSE).
– No dedicated “data/” folder, but data ingestion is implied in data_pipeline.R.
– No service or UI components—this is a one-off research batch job pipeline.
README for additional insights
• Single-year ML analysis, future multi-year extension.
• Data sourced from Pennsylvania ArcGIS endpoint.
• Methodological influences from a published paper.
• No mention of automated deployment or real-time service layers.