Medical service demand forecasting using a hybrid model based on ARIMA and self-adaptive filtering method

Area of Science:

• Healthcare Management
• Time Series Analysis
• Predictive Modeling

Background:

• Accurate forecasting of medical service demand is crucial for effective healthcare resource planning and allocation.
• Traditional time series methods like ARIMA are commonly used for short-term outpatient visit forecasting but have limitations in prediction horizon and accuracy.
• The inherent randomness, periodicity, and trend in daily outpatient volumes necessitate advanced forecasting approaches.

Purpose of the Study:

• To develop and validate a hybrid prediction model integrating ARIMA and a self-adaptive filtering method.
• To enlarge the prediction horizon and improve the accuracy of medical service demand forecasting.
• To compare the performance of the hybrid model against traditional ARIMA and other relevant models.

Main Methods:

• Utilized the Autoregressive Integrated Moving Average (ARIMA) model to identify time series features (cyclicity, trend) and estimate parameters.
• Employed a steepest descent algorithm within an adaptive filtering method to adjust ARIMA parameters and minimize prediction error.
• Validated the hybrid model using diverse datasets, including the Time Series Data Library (TSDL), a weekly emergency department (ED) case, and real-world data from a maternal and child health care center (MCHCC).

Main Results:

• The hybrid model demonstrated significant improvements in prediction accuracy, achieving 80-99% enhancement over ARIMA for TSDL cases.
• For a weekly ED visit case, the hybrid model outperformed traditional ARIMA and Artificial Neural Network (ANN) models.
• In real-world MCHCC data, the hybrid model drastically reduced Mean Absolute Percentage Error (MAPE) from 18.53% and 27.69% (ARIMA) to 2.79% and 1.25%.

Conclusions:

• The proposed hybrid prediction model significantly outperforms the traditional ARIMA model in forecasting accuracy, evidenced by smaller average relative errors.
• The hybrid model exhibits superior applicability for both short-term and medium-term medical service demand prediction.
• This approach offers a more reliable tool for healthcare resource planning and allocation.