Optimizing Rooftop Solar Potential Prediction: A Comprehensive Study Leveraging Machine Learning Techniques

Solar energy stands as a versatile and eco-friendly power source, pivotal in addressing global energy challenges and fostering a sustainable future. This study employs machine learning techniques to estimate solar energy potential in Manila, Philippines. Manual computations often yield inaccuracies and errors, highlighting the need for automated prediction methods. The heavy dependence on fossil fuels has greatly impacted the aforementioned sectors, which prompted the researchers to pursue this study. In order to fulfill its aim, an automated solar yield prediction has been implemented. The utilization of the Pearson Correlation Coefficient was done and employs a Backward Elimination Process to verify and confirm the results. The “All Features” has 10 features, and for the “Selected Features”, it amounts to a total of 7 features, with both categories having 4 dummy feature variables. A total of seven models, namely K-Nearest Neighbor, Ridge, Multiple Linear, Elastic Net, LASSO, Decision Tree, and Huber Regression were tested to evaluate their respective Mean Absolute Error, and R-squared scores, which enables the identification of differences between them. The model with the highest score for ${R}^{2}$ and lowest MAE will be chosen for implementation in the prototype. The validation processes were assessed using various metrics. Upon analysis and comparison, LASSO Regression was deemed the optimal choice for integration into the prototype, having $R^{2}$ score of 0.9987, with Mean Absolute Error of 761.3037, and aligning with the majority of the validation criteria.