Metaheuristic Approaches Integrated with ANN in Forecasting Daily Emergency Department Visits

Table 9

Comparison of previous studies from the literature.

Study	Data	Method(s)	Compared methods	Time frame	Variables	Performance measure	Results
Choudhury and Urena [6]	Data from January 2014 to August 2017	TBATS, Holt–Winters, neural net, and ARIMA	—	Hourly	Only temporal variables	ME and RMSE	1.75, 1.19, 1.40, and 1.00 (ME for each model, respectively); 2.28, 27.86, 3.26, and 1.55 (RMSE for each model, respectively)
Yousefi et al. [20]	Data from January 2014 to November 2016	Long short-term memory (LSTM)	MLR, ARIMA, SVR, GLM, GEE, SARIMA, and ARIMA-LR	Daily	Weekend, holiday, soccer match day, the day after holiday, and the day before holiday	MAPE and R-squared	5.55% (MAPE) and 0.940 (R-squared)
Zhang et al. [26]	Data from January 1, 2013, to December 31, 2016	ARIMA-SVR hybrid approach	ARIMA and SVR	Daily	Only temporal variables	MAPE, RMSE, and MAE	7.02% (MAPE), 19.20 (RMSE), and 14.97 (MAE)
Jilani et al. [23]	Data between Jan 2011 and December 2015 from four hospitals	A modified heuristics based on a fuzzy time-series model	ARIMA and ANN	Weekly and monthly	Only temporal variables	MAPE and RMSE	2.5% to 7% (MAPE-daily) and 2.09% to 2.81% (MAPE-monthly)
Khaldi et al. [14]	Seven years of aggregated weekly demand from 2010 to 2016	ANN with ensemble empirical mode decomposition (EEMD)	ANN with discrete wavelet transform (DWT) decomposition, ANN, and ARIMA	Weekly	Only temporal variables	RMSE, MAE, and R (correlation coefficient)	52.86 (RMSE), 39.88 (MAE), and 0.96 (R)
Tideman et al. [31]	Seven years of historical daily ED arrivals	Least absolute shrinkage and selection operator (LASSO) regression	—	Daily	Climatic variables, Google trends, and calendar variables	MAPE, RMSE, R-squared, and percent absolute percent error (PAPE)	7.58% to 10.99% (MAPE), 12.08 to 16.73 (RMSE), 0.13 to 0.57 (R-squared), and 3.29% to 11.23% (PAPE)
Carvalho-Silva et al. [22]	Data for ED arrivals in 2 years (2012-2013)	ARIMA	Moving average, multiplicative winters, Holt–Winters, and exponential smoothing	Daily	Only temporal variables	MAPE	5.92% to 10.63% (MAPE)
Sarıyer [11]	Data between 01/12/2016 and 28/02/2017	ARIMA	—	Daily	Only temporal variables	MAPE	5.01% to 8.16% (MAPE)
Juang et al. [15]	Monthly ED visits from January 2009 to December 2016	ARIMA	—	Monthly	Only temporal variables	MAPE	8.91% (MAPE)
Xu et al. [27]	Daily ED visits from January 1, 2012, to December 31, 2013	ARIMA-LR hybrid approach	GLM, ARIMA, ARIMAX, and ARIMA-ANN	Daily	Calendar, holiday and temperature variables	MAPE and RMSE	6.5% to 9.3% (MAPE of ED-1), 12.3% to 13.1% (MAPE of ED-2), 67.1 to 98.2 (RMSE of ED-1), and 5.49 to 5.73 (RMSE of ED-2)
Calegari et al. [28]	Period from January 1, 2013, to May 31, 2015	SS, SMHW, SARIMA, and MSARIMA	—	Daily	Calendar and climatic variables	MAPE	2.91% to 11.16% (MAPE)
Current study	Two years’ data from January 1, 2011, to December 31, 2012	Bayesian ANN, GA-based ANN, and PSO-based ANN	—	Daily	Calendar, holiday, and temperature variables	MAPE, MAE, MSE, RMSE, and R-squared	6% to 8.8% (MAPE), 40.888 to 60.358 (MAE), 2499.340 to 7031.078 (MSE), 49.993 to 83.852 (RMSE), and 0.343 to 0.824 (R-squared)