Unlocking solar potential through machine learning techniques for roof geometry prediction- A review

• Reviewed recent machine learning approaches for predicting residential roof geometry to support rooftop solar PV optimisation. • Introduced a dual-pillar review framework separating surface-normal estimation and roof geometry prediction to clarify methodological advances. • Compared analytical models with deep learning approaches such as U-Net and DeepLabv3+. • Classified available datasets by resolution, data modality (RGB, LiDAR, photogrammetry), and geographic coverage. • Highlighted key challenges in data quality, annotation consistency, and model generalisation across diverse urban contexts. The global transition to renewable energy underscores the necessity of optimising residential solar photovoltaic (PV) systems. Accurate roof geometry prediction is critical for enhancing the efficiency and scalability of rooftop solar installations. This review examines recent machine learning (ML) methodologies designed to enhance roof geometry prediction accuracy. It synthesises recent techniques, including Convolutional Neural Networks (CNNs) and multi-task learning frameworks, to evaluate improvements in predictive accuracy and computational efficiency. The review also highlights challenges such as data availability, diverse dataset integration (e.g., aerial imagery and LiDAR scans), and geographic generalisation, proposing opportunities for future research. Opportunities for future research include developing robust, scalable ML models and automated data preprocessing frameworks. Furthermore, the review advocates for accessible, open-source tools to accelerate the global adoption of sustainable solar energy. Overall, the findings aim to advance roof geometry predictions, improving the performance and broader implementation of residential solar PV systems.