Table 3. Forecast analysis.

Dengue estimation capacity of tweets up to 8 weeks in advance. The model selected (Table 2) was adjusted to different time lags between tweets and Dengue cases, the numbers (Tw_-1, Tw_-2,…Tw_-n) indicates the number of weeks that tweets are considered before the week of Dengue prediction.

MODEL	Model (Dengue ~ Negative Binomial (μt, k))	FORECAST (week in advance)	AIC	Deviance Explained	R-sq. (adj)	Mean relative error *
Twt	log(μt) = ƒ1(Tweetst) + ƒ2(weekt) + eyeart + β0	0	3805.52	93.7	0.94	0.34
Twt-1	log(μt) = ƒ1(Tweetst-1) + ƒ2(weekt) + eyeart + β0	1	3825.92	93.0	0.94	0.36
Twt-2	log(μt) = ƒ1(Tweetst-2) + ƒ2(weekt) + eyeart + β0	2	3849.33	92.2	0.92	0.40
Twt-3	log(μt) = ƒ1(Tweetst-3) + ƒ2(weekt) + eyeart + β0	3	3868.49	91.4	0.90	0.44
Twt-4	log(μt) = ƒ1(Tweetst-4) + ƒ2(weekt) + eyeart + β0	4	3877.07	91.0	0.89	0.45
Twt-5	log(μt) = ƒ1(Tweetst-5) + ƒ2(weekt) + eyeart + β0	5	3881.43	90.8	0.88	0.40
Twt-6	log(μt) = ƒ1(Tweetst-6) + ƒ2(weekt) + eyeart + β0	6	3896.20	90.1	0.89	0.40
Twt-7	log(μt) = ƒ1(Tweetst-7) + ƒ2(weekt) + eyeart + β0	7	3918.60	88.9	0.88	0.44
Twt-8	log(μt) = ƒ1(Tweetst-8) + ƒ2(weekt) + eyeart + β0	8	3931.54	88.2	0.87	0.46

* mean absolute relative difference between predicted Dengue cases and observed cases