DATA-DRIVEN FIRE RISK MANAGEMENT: SPATIO-TEMPORAL PREDICTION AND RESOURCE ALLOCATION MODELS
Madasseri Payyappalli, Vineet
MetadataShow full item record
Fire incidents, especially structure/building fires, are a highly common phenomenon. Structure fire incidents as impactful as the recent Notre Dame Cathedral fire of France may be rare, however, there are millions of fire incidents that happen in the U.S. every year. In 2014, there were more than 1.2 million fire incidents reported in the U.S., that had an overall economic cost of $328.5 billion, including about 3,300 deaths and 78,000 injuries. Taking advantage of various publicly available data sources, this research provides data-driven optimization and analytics models to assist decision making in fire risk management, to mitigate losses from fires. This research is primarily focused on the risk management of structure fires, however, comparisons with wildfire risk management are drawn as appropriate. This research work has three major chapters. The first chapter presents a detailed review of literature related to resource allocation, predictive modeling, and other analytics-based approaches in fire risk management. The areas that have not yet been explored in detail by operations research and analytics researchers are discussed. In addition, a comparative study of the models used in the contexts of structure fires and wildfires is presented. The second chapter investigates how strategic resource allocation decision making in fire risk management could be improved by using optimization techniques. First, an optimal resource allocation model (RA model) that features the trade-off between equity and efficiency, is presented along with numerical results and insights. A case study of federal fire grant allocation is used to validate and show the utility of the RA model. The results also identify potential underinvestment and overinvestment in fire protection in certain regions. The concept of fire risk scores for spatial and temporal units is explored. Additionally, this chapter investigates how the RA model can be extended to fire-department-level resource allocation. The equity considerations are made more need-based by incorporating fire risk scores calculated for counties and ZIP codes. The chapter concludes by summarizing challenges ahead of estimating parameters required for this model.The third chapter develops machine learning models that predict structural fire losses, at spatial and temporal levels. First, various data sets such as fire incident records, socio-economic/demographic data, fire department location and personnel data, and weather data, are merged. Second, feature engineering and feature selection techniques are used to process the data. Next, various machine learning models are trained and tested using this processed data set. The results from these models are compared, and a discussion is provided on using these models for drafting risk-informed funds allocation and pre-positioning of firefighting resources to mitigate fire risk.