Artificial intelligence can help forecast fires

– Using artificial intelligence and predictive modelling in forecasting accidents and incidents is a new and growing phenomenon. In recent years artificial intelligence has been used abroad to, among other things, forecast building fires, optimise the response times for rescue departments and to make forest fire and traffic accident forecasts, says Laura Kuurne, Researcher in Data Analytics at SPEK.

Predictive models have particularly been used to forecast building fires. Different data analysis tools that use their algorithms, such as Firebird, Firecast and Firecare, have been introduced in the USA.

– In practice, forecasting is done by humans ‘teaching’ the predictive model by entering data which includes detailed information about accidents, incidents, and buildings as well as socioeconomic information about the local population. The models are then tested by retro-forecasting accidents and incidents that have already happened. The percentage of success is determined by how many accidents which have already happened the model was able to predict using the human-taught algorithm. In prior case studies the models’ predictions have been encouragingly accurate, up to 65–80 per cent, says Ms Kuurne.

Data sourcing is a challenge. The models need as much information as possible regarding factors that may directly or indirectly contribute to the onset of accidents and incidents. Compiling the data is the responsibility of many actors and, due to data protection issues, it may be time-consuming and expensive. There may also be challenges in the quality and integrity of data which can, however, be partly corrected through artificial intelligence.

Take Finland’s rescue services risk analysis to the top international tier

The SPEK project studies whether the platforms introduced abroad could be applied in Finland as well, and whether similar techniques could be utilised in forecasting fires and other accidents and incidents. The project makes use of practical experiments by utilising the information at hand. The goal is to create an analytical tool that uses predictive models and supports the rescue services’ contingency planning, making it possible, for example, to suitably target monitoring on the basis of an algorithm-based risk analysis.

This project directly supports Work Package 2 of the Interior Ministry’s ongoing project Capacity of and planning criteria for rescue services and civil emergency preparedness. The work package reforms risk models and accident and incident predictions.

The goal is to create risk analysis methods and tools for nationwide use which the rescue departments could also utilise in practice. For example, the predictive model could be used for a one-off analysis or, when necessary, incorporated into a software application. If it is possible to update the data needed by the analysis in real time, it is also possible to develop the data tool into a version which generates real time risk analysis in support of rescue departments.

– In order to facilitate predictive and real time risk analyses data sourcing should be done in a more straightforward and centralised fashion for the rescue departments. For example, an information resource could be set up to serve rescue departments risk analyses. Its data would be updated in real time by, among others, the rescue services, the Finnish police, Statistics Finland, the Social Insurance Institution of Finland (KELA), and the Finnish Institute for Health and Welfare (THL), says Ms Kuurne.

– Such an information resource, coupled with a predictive modelling tool, would take Finland’s rescue services risk analysis to the top international tier.