Table 2. Tweets are a useful tool for estimating Dengue activity at country level.
Comparison between the selected model and other models with combinations of the variables: tweets, Dengue cases and temporal structures; using AIC, explained deviance and mean relative error as estimation capacity indicators.
| Explanatory Variables | Model (Dengue ~ Negative Binomial (μt, k)) |
AIC | Deviance Explained | Mean relative error * |
|---|---|---|---|---|
| Tweets + temporal structure + Dengue cases (three weeks lag) |
log(μt) = ƒ1(Tweetst) + Dent-3 + ƒ2(weekt) + eyeart + β0 | 3805.44 | 93.8 | 0.344 |
|
Tweets + temporal structure
(SELECTED MODEL) |
log(μt) = ƒ1(Tweetst) + ƒ2(weekt) + eyeart + β0 | 3805.52 | 93.7 | 0.345 |
| Temporal structure + Dengue cases (three weeks lag) |
log(μt) = ƒ(weekt) + eyeart + Dent-3 + β0 | 3917.47 | 88.5 | 0.442 |
| Temporal structure only | log(μt) = ƒ(weekt) + eyeart + β0 | 3948.18 | 86.6 | 0.510 |
| Tweets + Dengue cases (three weeks lag) |
log(μt) = ƒ(Tweetst) + Dent-3 + β0 | 4027.69 | 79.3 | 0.694 |
| Tweets only | log(μt) = ƒ(Tweetst) + β0 | 4103.40 | 69.7 | 0.954 |
| Dengue cases (three weeks lag) only | log(μt) = ƒ(Den t-3) + β0 | 4113.61 | 67.8 | 0.707 |
* mean absolute relative difference between predicted Dengue cases and observed cases