Data mining model for ambulance run forecast

Objective: The purpose of this study was to evaluate a forecast model to predict ambulance run and ambulance non-transport. Metorials and Methods: The model was set up by using a Back-Propagation Network for ambulance mission prediction. Input variables included month, holidays, time of ambulance call, result of ambulance call, temperature, and humidity. Output results included ambulance run and non-transport. Data from the Tainan City dispatch center from 1996 and 1997 were used for setting up the model. Data from 1998 was used to test the model. Results: The averages mean prediction error was between [-6.1, 6.1] percent. A larger error was noted during January and February (spring vacation period in Taiwan). Conclusion: The data mining model could provide forecasts of future ambulance service run volume and non-transport. With the development of medical informatics, using a predicting model for planning medical demand is feasible. Further evaluation and application are warranted for cost-effective medical resource management.