PREDICTING FIRE RISK USING MACHINE LEARNING AND REAL-TIME ACCESS CONTROL SYSTEM DATA: EVIDENCE FROM KIGALI SPECIAL ECONOMIC ZONE (KSEZ)

Abstract

This study investigates the application of machine learning techniques in predicting fire risk within industrial environments, using the Kigali Special Economic
Zone (KSEZ) as a case study. The research integrates environmental monitoring data and real-time access control system data to develop predictive models
capable of identifying fire risk levels. A dataset comprising 80,000 observations and 11 variables was analyzed using Python-based machine learning models,
including Logistic Regression, Random Forest, and XGBoost. The findings reveal that fire risk is not random but can be predicted using key indicators such as
equipment heat index, temperature, hazard level, and abnormal access patterns. Among the models tested, Logistic Regression achieved the highest accuracy of
99.81%, demonstrating strong predictive capability. The study further shows that integrating environmental and behavioral data significantly improves model
performance compared to single-source models.