Machine learning-based forecasting of urban fire impact in city environments

Area of Science:

• Disaster Management
• Artificial Intelligence
• Urban Safety

Background:

• Effective resource allocation is critical for fire departments to manage incidents efficiently.
• Predicting fire escalation can significantly improve response outcomes and reduce associated damages.

Purpose of the Study:

• To develop and validate a predictive model for estimating the likelihood of fire escalation.
• To inform fire department resource allocation strategies through data-driven insights.

Main Methods:

• Analysis of 47,382 fire incidents using an XGBoost model.
• Incorporation of building characteristics, temporal data, and GIS spatial features.
• Validation through 5-fold cross-validation, temporal holdouts, and geographic tests.

Main Results:

• The predictive model achieved 85.6% accuracy and an AUC of 0.83.
• Key predictors for fire escalation included older buildings, nighttime/weekend incidents, building structure, use, and floor count.
• Simulations indicated potential reductions of 25% in property damage, 21% in firefighter injuries, and 18% in response times.

Conclusions:

• Predictive analytics offers a powerful tool for enhancing real-time firefighting efficiency and public safety.
• The developed framework shows promise for optimizing resource allocation and incident management.
• Further validation in diverse urban settings and with refined severity scales is recommended for broader applicability.