Carnegie Mellon University

photo of a man inspecting a gauge

May 11, 2018

Fire Risk Analysis: Predicting Fire Risk to Prioritize Commercial Property Fire Inspections

Working with the Pittsburgh Bureau of Fire (PBF), the Fire Risk Analysis team is using historical fire incident and inspection data, coupled with business permits and property condition information to develop predictive models of structure fire risk for commercial properties. PBF conducts regular fire inspections of commercial properties, as stipulated by the municipal fire code. With a prioritization method informed by the prediction probability of fire risk from machine learning models, and implemented in an interactive map visualization, they will be better able to target their inspections for the properties at greatest risk of fire.

STATUS: Our team has developed and deployed a machine learning model to predict nonresidential structure fire risk with a predictive performance better than the current state-of-the-art. We trained the model on data of historical fire incidents, property inspections, and property assessments and evaluated multiple model types (e.g. random forests, SVM, XGBoost). The risk scores generated by the model are displayed in a data dashboard and interactive map used by fire inspectors and fire chiefs to inform the Bureau of Fire’s prioritization of property fire inspections, so they can inspect the properties at greatest risk of fire. The model is currently deployed on their servers and retrains on a regular basis, as new data is generated. In the months since the first risk scores were generated, 14 of the 45 building fire incidents (31%) occurred in one of the medium or high-risk properties, significantly higher than the 0.20% base rate for fire incidents in the city. Our future work includes incorporating new data sets, experimenting with new model types, and expanding this approach to predict fire risk in residential properties at the census block level. Our team is also working to make the model more easily extensible and usable by other cities’ fire agencies. 

A final report from this phase of the project can be found here.

Open-source code for this project can be found here.

This work was recently accepted to the KDD conference:

Singh, B., Hu, Q., Chen, J., Chen, F., Lee, J., Kuo, N., Narang, P., Batts, J., Arnold, G., Madaio, M. (2018). A dynamic pipeline for spatio-temporal fire risk prediction. In Proceedings of the 2018 ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) (in press). [pre-print pdf]


Pittsburgh Bureau of Fire

Pittsburgh Department of Innovation and Performance


Michael Madaio
Ph.D. Student, Human-Computer Interaction Institute, Carnegie Mellon University


Bhavkaran Singh
Qianyi Hu
Palak Narang
Nathan Kuo
Jeffrey Chen
Fangyan Chen
Jessica Lee