A unified machine learning approach to time series forecasting applied to demand at emergency departments

BMC Emergency Medicine

Table 5 MAE error rates for all types of hyperparameter tuning predicting hospital demand 1, 3 or 7 days in advance in case of the batch method for Charing Cross hospital

		Choosing the best set of hyperparameters based on:
algorithm	days	yesterday	the past	exponential	the average over the	caret
			n days	moving average	whole training set
	1	10.66	10.60	10.67	10.58	10.51
lm	3	10.75	10.72	10.84	10.69	10.69
	7	10.78	10.90	10.83	10.87	10.77
	1	12.50	10.68	11.12	10.68	10.89
gbm	3	12.49	10.58	10.92	10.50	10.78
	7	12.11	10.73	11.03	10.57	10.72
	1	10.72	10.51	10.59	10.51	10.51
glmnet	3	10.85	10.69	10.74	10.69	10.69
	7	10.86	10.76	10.84	10.76	10.76
	1	12.81	12.57	12.62	12.77	12.36
knn	3	12.85	12.58	12.78	12.87	12.51
	7	12.65	12.45	12.41	12.47	12.33
	1	11.09	10.84	11.04	11.55	10.89
rf	3	10.98	10.80	10.83	11.85	10.78
	7	10.92	10.78	10.79	11.99	10.72

ISSN: 1471-227X