The global burden of dengue, a mosquito-borne viral infection, has alarmingly increased in recent decades. The rise in disease occurrence is mainly attributed to changes in the climate, human ecology, globalization, and demography. In such a scenario, an accurate prediction of a dengue outbreak is essential to reduce the morbidity rate significantly. Therefore, this paper employs two classes of autoregressive models for dengue forecasting and a recently proposed approach called Finite Element Machine for Regression (FEMaR). Further, it proposes a variant of the latter, namely FEMaR-KD, which allows the exploration of k -approximate nearest neighbors to interpolate data points based on k -neighborhood instead of the whole dataset. Such models are built considering environmental parameters, which denote one of the main determinants for infection occurrence. Finally, the proposed models’ performance is assessed over two distinct datasets, considering differing spatial scales and regions. Results show that FEMaR obtained Mean Absolute Error up to 51% smaller than the autoregressive models considering univariate scenarios and Root Mean Squared Error up to 63% smaller regarding the univariate models.
