The Dutch fire and rescue services are developing an interest in the use of Business Intelligence for their operations in order to improve safety, have tighter financial control, and better risk management. To this end, a rich database of incidents has been created. In this project (NWO Applied and Engineering Sciences Open Technology Programme 18004), we develop advanced probabilistic models and a rigorous statistical toolbox to obtain operationally usable predictions and quantify uncertainty.
We considered a number of different types of fires:
- for chimney fires, a nested Poisson model was developed and is now being implemented for operational use by EngD candidate Ömer Esas,
- for kitchen fires, a non-parametric tree-based model was proposed,
- a Bayesian hierarchical model is being developed for arson fires.
This research is carried out in close cooperation with the Twente Fire Brigade (TFB).
Results
M.L. School, M. de Graaf, M.N.M. van Lieshout E.M.A. Sanders, R.A.C. de Wit (2021). Van tellen naar voorspellen. Sturen op risico’s met een voorspellend wiskundig model op basis van historische brandweerdata. Tijdschrift voor Veiligheid 20:60-74.
C. Lu, M.-C. van Lieshout, M. de Graaf, P. Visscher (2021). Chimney fire prediction based on environmental variables. Bulletin of the International Statistical Institute LXIII (Proceedings ISI World Statistics Congress 2021).
M. de Graaf, M.-C. van Lieshout, R.A.C. de Wit (2021). From tally to foretelling. Interview in I/O Magazine, pages 19--21, October 2021.
M.N.M. van Lieshout, C. Lu (2022). Infill asymptotics for logistic regression estimators for spatio-temporal point process intensity estimation. Arxiv 2208.12080.
D. van Leeuwen (2022). Data-driven kitchen fire prediction based on environmental variables. BSc thesis, University of Twente.
C. Lu, M.N.M. van Lieshout, M. de Graaf, P.J. Visscher (2023). Data-driven chimney fire risk prediction using machine learning and point process tools. Annals of Applied Statistics, 17:3088–3111.
M.N.M. van Lieshout, C. Lu (2023). Contribution to the discussion of “Automatic change-point detection in time series via deep learning” by Li et al. Journal of the Royal Statistical Society, to appear.
C. Lu, Y. Guan, M.N.M. van Lieshout, G. Xu (2024). XGBoostPP: Tree-based estimation of point process intensity functions. ArXiv 2402.17966.
L.N. van Kasteren (2024). Chimney fire prediction in IJsselland. BSc thesis, University of Twente.
Researchers
Maurits de Graaf, Marie-Colette van Lieshout, Changqing Lu, Ömer Esas, Paul Visscher
Students: Dorien van Leeuwen, Lars van Kasteren