A Comparison of Extreme Gradient Boosting and Convolutional Neural Network-Long Short-Term Memory for Service Demand Forecasting

Abstract

The accurate demand forecasting gives powerful insights on amount of the available resources should be allocated for the operation and can improve the effectiveness of the management to assure the completeness of the services. This work studied to forecast the number of patients, that applied for service demand forecasting. We proposed the comparison between XGBoost model and CNN-LSTM model for service demand forecasting. We created the XGBoost model and configured the optimal parameters by using grid search optimization techniques. In the same way, the CNN-LSTM model was constructed and adjusted the layers in the network structure. Later, the model has configured the optimal parameters obtained from the grid search optimization techniques. Both models were applied to two different datasets for forecasting the number of patients in the future. The results indicated that two models had skillful, and made the reliable forecasting in two datasets. This work evaluated and measured the model performance by calculating RMSE, and sMAPE, which was acceptable in all case study. The CNN-LSTM model gave better efficiency forecasting than XGBoost model. In contrast, XGBoost model operated faster, took a lower performing time, and lower CPU consumption than the CNN-LSTM model.