Electricity Demand Forecasting System

Developed a machine learning pipeline to forecast electricity demand by combining Zenodo energy datasets with ERA5 weather data. Engineered advanced features such as lagged variables, rolling averages, and Cooling Degree Days (CDD) to capture heatwave effects. Evaluated models including Random Forest and XGBoost, achieving high accuracy (MAE ~2.1, R² ~0.998). Key insights showed extreme heat as the main driver of prediction errors, highlighting the importance of weather-aware forecasting for grid stability.

Tech stack

• Python
• Scikit-learn
• XGBoost
• Pandas
• Matplotlib
• NumPy