Modeling and prediction fire evolution in Building engineering: a knowledge graph and machine learning approach

Building engineering fire risk management faces growing challenges owing to rapid urbanisation and industrialisation. In dense urban areas, high-rise clusters, multi-functional zones, and interconnected infrastructure increase the complexity of fire spread, posing major threats to public safety. Accurate prediction of fire evolution is essential for effective risk mitigation. However, current methods often rely on subjective and static approaches, which are inefficient and lack scalability. Integrating knowledge graph with machine learning improves prediction accuracy and practical applicability. This study analyses 510 Chinese building engineering fire accidents (2000–2024), constructing a knowledge graph to integrate diverse urban data. Three models, logical convolutional neural network (L-CNN), random forest, and k-nearest neighbours, were applied for fire evolution prediction. The L-CNN model achieved the highest accuracy (83.29%) and lowest variance (0.27), demonstrating superior adaptability in complex fire scenarios. Further sensitivity analysis was conducted to rigorously assess the robustness of the predictive model outputs. This research supports early warning systems and emergency decision-making, advancing data-driven fire safety management.