Energy Demand & Generation Forecasting Hub
Comparative regression pipelines (Random Forest, XGBoost) in Python and R with modular continuous training, served via FastAPI and Streamlit.
Project Overview
Modular system for forecasting electrical energy generation and demand using historical multivariate time-series data. Monolithic legacy scripts were refactored into a clean, testable package structure with comprehensive comparative benchmarks across generalized linear models, Random Forest, and XGBoost algorithms.
Key Milestones
- Model Performance Benchmarking: Selected Random Forest for production deployment after outperforming alternative baselines on Root Mean Squared Error (RMSE) metrics.
- Modular Pipeline Architecture: Automated continuous feature engineering for cyclical seasonality and multi-step time horizons.
- Microservice & Dashboard: Containerized inference API built with FastAPI and Docker, paired with an interactive monitoring and exploration dashboard in Streamlit.