Forecasting Inpatient and Outpatient Visits for Depressive Disorders: A Comparative Study of Deep Learning Approaches

Abstract

Forecasting inpatient and outpatient visits is essential for successful resource allocation and clinical decision-making. The techniques used by previous researchers for forecasting, primarily based on statistical approaches, which often require extensive data preprocessing and expert knowledge, can be time-consuming and difficult. Therefore, this study analyses three current deep learning (DL) algorithms, recurrent neural network (RNN), long short-term memory (LSTM) and gated recurrent unit (GRU), for forecasting inpatient and outpatient visits for depressive disorders. These algorithms are among the most familiar DL techniques for time series and have been used with remarkable success in various contexts. The DL algorithms were evaluated using mean squared error (MSE), root mean squared error (RMSE) and mean absolute error (MAE). Based on the results comparison, LSTM has the best performance (lowest error values) compared to the RNN and GRU. The DL algorithms are also being compared to state-of-the-art algorithms, and the results show that the DL algorithms can accurately forecast inpatient and outpatient visits compared to the previously proposed algorithms. The findings from this study could be helpful in clinical decision-making and resource allocation in mental health care.

Keywords

• deep learning
• depression
• psychiatric department
• time series forecasting