Prediction of emergencies

Important results from the project

The aim of this study is to investigate if it is possible to enhance the preparedness model that is used today based on data from SOS Alarm, SCB and OpenStreetMap. Today´s preparedness model bases its decisions mainly on case history from SOS Alarm. The goal of this study is to ascertain if one could obtain an even better forecast of where the emergencies (priority 1 and 2) will take place, with the help of machine learning techniques that are based on temporal and spatial data, rather than using a model that is based on a moving average.

Expected long term effects

The study presents a proposal of a new preparedness model that could replace the existing model. The model that is presented showed a true positive rate (TPR) of 97 % and a true negative rate (TNR) of 59 % which is an improvement compared to the existing model whose TPR and TNR show 99,5 % and 29,8 respectively. The result turns out to be better at correctly identify more emergencies at the expense of more incorrectly predicted non-emergencies, which means that the proposed model has the potential to replace the existing preparedness model.

Approach and implementation

The study was divided in four parts: Planning, pre-study, implementation and writing the report. After first planning to create multiple models that covers the whole of Sweden, it was finally chosen to create model over Västra Götaland, since the problem to solve would be the same. Different types of approaches were explored during the pre-study including model architecture and how to model the data. In the last step a report about the study was written where the different approaches and the result was presented.