How to Build an Optimal and Operational Knowledge Base to Predict Firefighters’ Interventions

Abstract

It has recently been shown that pre-emergency transport, whether performed by firefighters or private ambulances, has a predictive character due to the fact that rescue is directly related to human activity, which is itself predictable. XGBoost has emerged as the best tool to predict the number of interventions by type, but how to design an optimal and operational knowledge base has not been discussed so far. We propose to explain how to make such a base with a content that is both relevant and can be continuously updated, making possible the industrialization of the process, and thus a better operational response of the concerned services. We show that three feature selection tools custom-built for XGBoost are mature enough to allow the optimization of such a database, and a good accuracy in predictions. We also show what these tools can bring in terms of business knowledge, and discuss the organizational and efficiency consequences that such an optimized predictive model could bring.

Predictions in Pre-Hospital Emergency Transport in France: A State of the Art

“…XG Boost has emerged as the best tool to predict the number of interventions by type, but how to design an optimal and operational knowledge base has not been discussed until now. In [9], we explain how to realize such a base with a con-tent that is both relevant and can be continuously updated, making possible the industrialization of the process, and thus a better operational response of the concerned services. We show that three feature selection tools custombuilt for XG Boost are mature enough to allow the optimization of such a database, and a good accuracy in predictions.…”