Comprehensive Building Fire Risk Prediction Using Machine Learning and Stacking Ensemble Methods

Abstract

Building fires pose a critical threat to life and property. Therefore, accurate fire risk prediction is essential for effective building fire prevention and mitigation strategies. This study presents a novel approach to predicting fire risk in buildings by leveraging advanced machine learning techniques and integrating diverse datasets. Our proposed model incorporates a comprehensive range of 34 variables, including building attributes, land characteristics, and demographic information, to construct a robust risk assessment framework. We applied 16 distinct machine learning algorithms, integrating them into a stacking ensemble model to address the limitations of individual models and significantly improve the model’s predictive reliability. The ensemble model classifies fire risk into five distinct categories. Notably, although the highest-risk category comprises only 22% of buildings, it accounts for 54% of actual fires, highlighting the model’s practical value. This research advances fire risk prediction methodologies by offering stakeholders a powerful tool for informed decision-making in fire prevention, insurance assessments, and emergency response planning.

Keywords

• fire risk prediction
• building fires
• machine learning algorithms
• stacking ensemble model
• fire prevention