Author: Billy Pierre
Tools: Python, Pandas, Matplotlib, Seaborn, Scikit-learn, WordCloud
Platform: Google Colab
Analyze over 400 industrial accident records to identify patterns in severity, risk types, sectors, and timing — and build a predictive model for high-severity incidents.
- Source: Kaggle – IHMStefanini Industrial Safety Dataset
- Records: 400+
- Fields: Date, Sector, Gender, Role, Severity, Critical Risk, Description
- Most accidents were low severity (Level I), but many had high potential (Levels IV–VI)
- Pressed, Manual Tools, and Chemical Substances were top risks
- Mining and Metals were the most incident-prone sectors
- Peak accidents occurred in early months (Feb–June)
- Third-party and male workers had higher incident rates
A Random Forest classifier predicted high-potential severity incidents using:
- Accident Level
- Industry Sector
- Gender
Model outputs include a confusion matrix and feature importance chart.
This project includes the following key visuals:
- Top Critical Risks (with/without Other/Unknown)
- Heatmap: Severity by Industry
- Monthly Trends Line Chart
- Word Cloud of Incident Descriptions
- Confusion Matrix & Feature Importance Chart
pip install -r requirements.txt- Expand text mining on incident descriptions
- Deploy as real-time safety dashboard
- Use model insights to build alert systems
For feedback or collaboration, feel free to connect with me on LinkedIn.