基于ARIMA-GRNN组合模型预测医保补偿住院费用

目的 探讨ARIMA-GRNN组合模型在城乡居民基本医疗保险的医保补偿住院费用预测中的应用，比较它与ARIMA模型预测的效果。方法 收集2016年1月至2020年12月河南某市城乡居民基本医疗保险的住院补偿费用，使用Matlab R2016a建立ARIMA-GRNN组合模型，使用R 4.0.3建立ARIMA模型，用2020年5月至12月的数据对两模型的预测效果做评价。结果 ARIMA(0,1,1)(1,0,0)12模型相对最优，其拟合部分的MAE、MAPE、RMSE分别为3998.4、10.9%、5642.9，预测部分的MAE、MAPE、RMSE分别为6521.9、16.5%、6675.8。ARIMA-GRNN模型最优光滑因子为0.166，其拟合部分的MAE、MAPE、RMSE分别为4044.5、11.1%、5622.0，预测部分的MAE、MAPE、RMSE分别为5831.8、14.8%、6013.7。结论 ARIMA-GRNN组合模型预测效果比单纯ARIMA模型好，可短期预测医保补偿住院费用。

Prediction of hospitalization expenses compensated by medical insurance based on ARIMA-GRNN hybrid model

To explore the predictive effect of ARIMA-GRNN hybrid model in the prediction of hospitalization expenses compensated by medical insurance of urban and rural residents and compare it with ARIMA model. Methods Based on the hospitalization expenses compensated by medical insurance of urban and rural residents in a city of Henan Province from January 2016 to December 2020, ARIMA model was established with R 4.0.3, ARIMA-GRNN hybrid model was established with MATLAB R2016 a, and the prediction effect of the model was evaluated with the data from May to December 2020. Results The MAE, MAPE, and RMSE of the fitting part of the ARIMA(0,1,1)(1,0,0)12 model were 3 998.4, 10.9%, and 5 642.9, respectively. The MAE, MAPE, and RMSE of the prediction part were 6 521.9, 16.5%, and 6 675.8. The optimal smoothing factor of the ARIMA-GRNN model was 0.166. The MAE, MAPE, and RMSE of the fitting part were 4 044.5, 11.1%, and 5 622.0, and the MAE, MAPE, and RMSE of the prediction part were 5 831.8, 14.8%, and 6 013.7. Conclusion The predictive effect of ARIMA-GRNN hybrid model is better than that of ARIMA model alone, so the hybrid model can be used for short-term prediction of hospitalization expenses compensated by medical insurance.