Barbados Traffic Analysis Challenge
ML solution for predicting traffic congestion using video analysis. Implemented YOLO detection, VideoMAE embeddings, and ensemble models.
Machine Learning · Data Science · Energy · XGBoost
Scikit-learn, Pandas, NumPy, Data Visualization
★ 5 stars on GitHub LinksCode All projects
Overview
Led a team to 2nd place at the DataTour 2024 - pan-African Data Science Nations Cup, a competition featuring 100+ teams from across Africa. The challenge focused on predicting energy demand to improve resource allocation across the continent.
The Challenge
- Objective: Forecast energy demand across 50+ African countries
- Data: Socio-economic and environmental factors including GDP, population, climate data, and historical energy consumption
- Evaluation: Based on prediction accuracy and model interpretability
Our Approach
Data Analysis
- Extensive exploratory data analysis to understand patterns across different regions
- Feature engineering based on domain knowledge of energy consumption drivers
Modeling
- Ensemble of regression models (Gradient Boosting, Random Forest, XGBoost)
- Cross-validation strategies accounting for geographical and temporal dependencies
- Feature importance analysis for interpretability
Results
- Ranked Top 3 on the final leaderboard
- Strong performance on both accuracy and robustness metrics
Technologies
- Data Processing: Pandas, NumPy
- Visualization: Matplotlib, Seaborn
- ML: Scikit-learn, XGBoost, LightGBM
Impact
Accurate energy demand forecasting helps African nations better plan infrastructure investments and resource allocation, contributing to sustainable development goals.