Prediction of urban medical emergencies using machine learning models based on spatial and temporal variables

Area of Science:

• Urban planning
• Public health
• Data science

Background:

• Accurate forecasting of medical emergencies is crucial for optimizing emergency services.
• Effective resource allocation can improve response times and reduce mortality.
• Urban emergency prediction models are needed for preparedness.

Purpose of the Study:

• To develop and evaluate a prediction model for medical emergencies in the Bucaramanga metropolitan area.
• To enhance preparedness and optimize resource allocation for medical emergency services.
• To reduce response times, prevent injuries, and lower mortality and morbidity rates.

Main Methods:

• A random forest model with sliding window techniques was employed for automatic emergency prediction.
• Temporal variables, spatial locations, meteorological data, road conditions, demographics, and traffic statistics were integrated.
• Two datasets were used: Regulatory Center for Emergencies and Urgencies (RCEU) records and Bucaramanga traffic accident reports (2017-2019).

Main Results:

• The model achieved an MSE of 0.005 with the RCEU dataset and 0.018 with traffic accident reports.
• Aggregated daily spatial predictions showed promising performance with an of 0.75 (RCEU) and 0.898 (traffic department).
• The sliding window approach effectively identified periods of increased emergency occurrences.

Conclusions:

• The automated prediction model shows promise in identifying periods with higher emergency occurrences across a 165 km² urban area.
• While fine-grained accuracy is limited, the model aids in potential injury prevention strategies.
• Further advancements using fine-grained data and machine learning techniques could enhance prediction precision.