This research presents a complete machine learning approach for photovoltaic (PV) energy prediction utilizing operational data from various Politecnico di Torino installations. Renewable energy integration requires accurate PV estimates for grid stability, energy management, and cost optimization. Data analysis, robust preprocessing using feature engineering, statistical pattern recognition, and advanced machine learning model creation are used to analyze five 2014-2024 datasets. This involved testing linear regression, random forest, and XGBoost with systematic hyperparameter optimization in single-site, multi-site, and directed east-west configurations. Results show that advanced machine learning approaches accurately forecast real-world PV installations (R² = 94-96%). Random Forest performed well, while XGBoost excelled in extreme conditions.