Projects
Road Accidents in France
Motivation
Each year, about 1.3 million people die from road accidents (source: WHO). The goal of this Data Science project is to investigate the influences on the severity and occurrence of traffic accidents. The knowledge gained should help political decision-makers to take measures which may prevent lethal traffic accidents.
Info
This is the repository for my personal data science project on road accidents in France. This is a work in progress; have a look at the repository to check the current status and more information.
The data is provided by the French government.
My Work
- import raw data and clean it
- visualize a number of relationships between variables
-
predict the severity of road accidents with means of
- conventional machine learning (XGBoost, Random Forest with Bayesian Search), and
- neural networks (Tensorflow Keras)
- interpret the applied models (Permutation feature importance) .
- organize the project with git, using GitHub actions to automate linting with Pylint (ensure compliance with the PEP 8 style guide) and testing (with papermill).
Backlog
- Interpretation of Artificial Neural Networks (with permutation feature importance)
- Analysis of the occurence of accidents with means of time series analysis (statsmodels)
Findings
The permutation feature importance for XGBoost shows, that the severity (3 classes: unharmed, injured, killed) of accidents is largely affected by variables of place and time […]
Used Libraries and Tools (Selection)
- Python3
- Pandas
- scikit-learn
- scikit-optimize
- Seaborn
- Git
- Jupytext
- Bokeh
- GitHub Actions
- Pylint
- XGBoost
- TensorFlow
- Keras
- Makefile
- Streamlit
View Jupyter Notebooks
Badges
Links
Kaggle Projects
Other Coding Projects (not ML-Related)
Kay Langhammer 2024