GC SafeSite AI — Construction-Site Risk Prediction
- Date : 2026
- Artificial Intelligence
- Completed
Built an AI system that predicts construction-site safety risks from operational data and inspection reports, with a real-time decision dashboard.
Overview
Data mining and artificial intelligence project carried out for Carthage Assurances: an intelligent system that predicts safety risks on construction sites from operational data and textual inspection reports.
The goal was to detect, analyse and classify risky situations so that decision-makers can anticipate accidents, manage on-site safety more effectively and strengthen prevention.
The system is built for safety managers, site supervisors and insurers: it brings artificial intelligence directly into a concrete decision-making process.
Key figures
- risk levels classified (low → critical)
- 4
- risk levels classified (low → critical)
- model families compared
- 5
- model families compared
Technologies
Python
pandas
Scikit-learn
XGBoost
Matplotlib
Seaborn
Streamlit
NLP
NLTK
spaCy
MLPProblem
Construction-site accidents are costly for workers, companies and insurers, yet the warning signs are scattered across operational data and free-text inspection reports that are hard to exploit manually.
Solution
A complete data mining pipeline — exploratory analysis, cleaning, feature engineering, class-imbalance handling (SMOTE, oversampling), then comparison of logistic regression, random forests, SVMs, neural networks and ensemble models — wrapped in an interactive Streamlit dashboard that scores a site’s risk level and raises alerts.
My contribution
Designed the end-to-end pipeline: data exploration and cleaning, feature engineering, model training, evaluation and optimisation, interpretation and comparison of results, then built the Streamlit dashboard.
Results
The final model delivers high accuracy and reliable classification of risk levels, with alerts and recommendations that make it a practical decision-support tool. With the IIT-Team, the project won 2nd place at the “Leveraging AI in Civil Engineering” competition organised by ISET Sfax.
Features
- Risk classification into four levels: low, medium, high, critical
- Combined use of structured site data and textual inspection reports
- Class-imbalance handling with SMOTE and oversampling
- Benchmark of several models (logistic regression, random forest, SVM, neural networks, ensembles)
- Interactive Streamlit dashboard to simulate real-world scenarios and assess a site’s risk
- Real-time alerts and safety recommendations to support decision-making