Machine Learning for Ambulance Demand Prediction in Stockholm County : Towards efficient and equitableDynamic Deployment Systems
Abstract: Pre-hospital care is a widely discussed subject with many actors working on figuring out what factors determine the outcome for the patient and how those factors can be affected. One factor believed to have a major impact on patient outcome is ambulance response time. A proposed way to improve response time is dynamic deployment systems. These systems require detailed predictions of spatio-temporal ambulance demand in order to function effectively. The purpose of the study is to explore the possibility of using machine learning to build a high-resolution predictor that dynamic deployment systems can use to reduce response time. In this paper we first try out unsupervised machine learning algorithms to dividing Stockholm County into small subregions (clusters) over which predictions can be made. Then, based on the best cluster-structure obtained, we train and evaluate a logistic regression model an to make probabilistic predictions of ambulance demand over these clusters. We compare it to a baseline model and although the logistic regression model outperforms the baseline in total, it is worse at predicting when dispatches actually happens. Either way, we see that risk-terrain data and historic dispatches data seem to be useful for predicting ambulance demand. In the end of this paper we evaluate how suitable today’s key performance indicators (KPI:s) are for Emergency Medical Systems(EMS) implementing dynamic deployment. We find that these systems likely entail a need for updated KPI:s for measuring effectiveness and equity.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)