Accurate Solar Radiation Forecasting Through Feature-Enhanced Decision Trees and Wavelet Decomposition

This study explores the efficacy of the decision tree algorithm in predicting solar power generation, addressing the inherent variability in photovoltaic (PV) energy production due to weather conditions. The main research question investigates how well decision tree models, enhanced by wavelet transform, can forecast daily solar radiation. Using data from NASA (January 2023 to April 2024), the study compares model performance with and without additional environmental features (e.g., temperature, clearness index, pressure). Key findings indicate that integrating these features significantly enhances predictive accuracy, reducing the Mean Squared Error (MSE) from 1.54 to 0.39 and the Root Mean Squared Error (RMSE) from 1.24 to 0.62. Further improvement is achieved with wavelet decomposition, lowering MSE to 0.02, RMSE to 0.15, and Mean Absolute Percentage Error (MAPE) to 3.12%, while increasing the R² score to 0.99. These results underscore the potential of wavelet-transformed decision tree models in accurately forecasting solar radiation.