Prediction analysis of the number of pre-hospital emergency ambulance trips in Handan based on the LPro Ensemble Model
10.3760/cma.j.cn114656-20250421-00307
- VernacularTitle:基于LPro集成模型的邯郸市院前急救出车车次预测分析
- Author:
Feng TIAN
1
;
Chengcheng BI
;
Penghui LI
;
Haifang ZHANG
;
Tingting ZHAO
;
Zhenjie YANG
;
Xian WANG
;
Jiaxuan GU
;
Shitao ZHOU
;
Zengjun JIN
;
Zhen WANG
;
Feifei ZHAO
;
Xianhui SU
;
Longqiang ZHANG
;
Saicong LU
Author Information
1. 河北工程大学，邯郸 057250
- Keywords:
Time series model;
LPro ensemble model;
Pre-hospital emergency care;
Trend forecasting
- From:
Chinese Journal of Emergency Medicine
2025;34(11):1530-1537
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the application of time series models in forecasting pre-hospital emergency ambulance trips in Handan City and develop the LPro ensemble model for improved prediction accuracy to support emergency resource allocation.Methods:Pre-hospital emergency data from Handan Emergency Medical Command Center (2019-2023) were retrospectively analyzed. From 324 799 original records, 289 949 valid records were included after cleaning. The training set (2019-2022: 215 918 records) included 35 527 records in 2019, 52 015 in 2020, 61 836 in 2021, and 66 540 in 2022. The validation set (2023) contained 74 031 records. ARIMA, linear trend seasonal, exponential smoothing, and Prophet models were fitted to the training set. The LPro ensemble model was constructed using MAPE-based weighting (linear trend seasonal model: 0.38, Prophet: 0.62). Performance metrics included MAPE, RMSE, MAE, and R 2. Results:Data showed annual growth (compound annual growth rate 23.27%) and seasonal patterns (October peaks, February troughs). Ambulance dispatches increased annually with monthly cyclical patterns. For 2023 validation predictions: ARIMA (MAPE 8.76%, RMSE 619, MAE 491, R 2 0.4563), linear trend seasonal (MAPE 9.83%, RMSE 671, MAE 545, R 2 0.3608), Prophet (MAPE 8.43%, RMSE 562, MAE 503, R 2 0.5513), exponential smoothing (MAPE 8.08%, RMSE 643, MAE 410, R 2 0.4124). LPro model showed superior performance (MAPE 7.05%, RMSE 491, MAE 393, R 2 0.6570), with 16.37% lower MAPE, 12.63% lower RMSE, 21.87% lower MAE, and 19.17% higher R 2 versus Prophet. Conclusion:The LPro ensemble model substantially enhances prediction accuracy and reliability, offering scientific support for emergency resource optimization and dispatch scheduling in Handan City.
