Final project for a Safety Management course, covering two domains:
- Occupational incidents in toy factories in China (binary outcome: accident vs. no-accident).
- Drunk-driving counts on road segments (count outcome by segment).
The work includes EDA, feature exploration, logistic regression, Poisson vs. Negative Binomial modeling with overdispersion diagnosis, and performance evaluation (ROC/AUC, sensitivity thresholds). All answers are organized by Q1, Q2, … matching the assignment.
project.ipynb– Assignment instructions (reference).safety_analytics_project.ipynb– My full solutions (answers by Q1, Q2, …).
- EDA & Visualization: histograms, scatter matrices, bar charts.
- Classification (Part 1): Logistic regression, coefficients interpretation, baseline probabilities, ROC/AUC, sensitivity-driven thresholding.
- Counts (Part 2): Linear regression sanity checks → Poisson GLM → Negative Binomial when overdispersion is detected.
- Model Diagnostics: constant-variance checks, residual patterns, AIC/BIC and log-likelihood comparison, overdispersion parameter, feature removal tests (AUC impact).
- Interpretability: odds/odds-ratio, IRR, top drivers by features (area, time-of-day, categories).
| File/Folder | Description |
|---|---|
project.ipynb |
Official assignment instructions (reference notebook). |
safety_analytics_project.ipynb |
My full solution notebook with answers (Q1, Q2, …). |
df_task_1_group_25.pkl |
Dataset for Part 1 – toy factories accidents (binary classification). |
drunk_driver_grpoup_25.pkl |
Dataset for Part 2 – drunk-driving counts (count regression). |
README.md |
Project documentation and overview. |
You can get this project in two ways:
Option 1 – Using Git
git clone https://github.com/adirbella37/safety-analytics-project.git
cd safety-analytics-projectOption 2 – Download as ZIP
- Click the green Code button at the top of this repository
- Select Download ZIP
- Extract the ZIP file on your computer
This project is licensed under the MIT License.