Table 4.

Accuracy comparison of multi-step-ahead LSTM modeling with two groups of input features and ARIMA using root-mean-square error (RMSE) and mean absolute error (MAE). The two indices were computed based on the actual and predicted weekly changes in natural log-transformed dengue cases.

Model			2018–2019		2019 Peak Period
			RMSE	MAE	RMSE	MAE
LSTM modeling	LSTM with NDVImean, RHmean, Rsum, and Tmean	1-week	0.36	0.29	0.28	0.23
		2-week	0.35	0.28	0.30	0.23
		3-week	0.36	0.28	0.34	0.26
		4-week	0.32	0.25	0.22	0.18
		5-week	0.36	0.29	0.29	0.24
		6-week	0.36	0.29	0.31	0.25
		7-week	0.38	0.3	0.35	0.29
		8-week	0.37	0.29	0.36	0.28
		9-week	0.38	0.3	0.34	0.29
		10-week	0.36	0.29	0.34	0.27
		11-week	0.36	0.29	0.34	0.29
		12-week	0.36	0.27	0.31	0.25
	LSTM with historical dengue data, NDVImean, RHmean, Rsum, and Tmean	1-week	0.35	0.27	0.23	0.20
		2-week	0.34	0.27	0.22	0.19
		3-week	0.34	0.27	0.25	0.20
		4-week	0.35	0.26	0.25	0.21
		5-week	0.34	0.27	0.22	0.19
		6-week	0.40	0.31	0.26	0.21
		7-week	0.37	0.30	0.28	0.22
		8-week	0.38	0.29	0.29	0.23
		9-week	0.38	0.29	0.32	0.27
		10-week	0.39	0.31	0.28	0.22
		11-week	0.34	0.27	0.28	0.23
		12-week	0.40	0.33	0.33	0.28
Baseline	ARIMA (3, 1, 2)		1.60	1.18	2.68	2.51