FWin transformer for dengue prediction under climate and ocean influence

TL;DR

This paper tackles long-range dengue forecasting under climate and ocean influences by developing a Fourier Mixed Window (FWin) transformer that blends local window attention with Fourier-based token mixing to handle non-stationary time-series. The authors demonstrate that FWin achieves superior predictive accuracy on Singapore dengue data (2000–2019) for horizons up to 60 weeks, especially when future predictor information is incorporated through the Modified Multivariate to Univariate task. Key contributions include a comprehensive comparison of transformer-based models, an ablation study on window size and input length, and evidence that cross-window mixing via Fourier transforms enhances long-range forecasting. The work has practical implications for public health planning by enabling more reliable, extended dengue projections under varying climate conditions and oceanic influences, with explicit avenues for future spatial-temporal extensions and domain-informed modeling layers.

Abstract

Dengue fever is one of the most deadly mosquito-born tropical infectious diseases. Detailed long range forecast model is vital in controlling the spread of disease and making mitigation efforts. In this study, we examine methods used to forecast dengue cases for long range predictions. The dataset consists of local climate/weather in addition to global climate indicators of Singapore from 2000 to 2019. We utilize newly developed deep neural networks to learn the intricate relationship between the features. The baseline models in this study are in the class of recent transformers for long sequence forecasting tasks. We found that a Fourier mixed window attention (FWin) based transformer performed the best in terms of both the mean square error and the maximum absolute error on the long range dengue forecast up to 60 weeks.

Figures (5)

• Figure 1: Sample features of the dataset. The plot includes the average temperature and precipitation from 2000 to 2019 with the dengue cases number. Here blue, orange, green indicate training, validation and testing split respectively.
• Figure 2: (a) Input-Output structure of MS task. (b) Input-Output structure of MM task. Here orange shaded cells indicate non-zero value of the response feature, blue shaded cells indicate non-zero value of the predictor features, and white cells indicate zero padded value.
• Figure 3: FWin model overview fouriermixedtran2023.
• Figure 4: Sample VAR predictions with various lag orders. The $x$-axis represents time in weeks, and $y$-axis is the normalized cases number. The suptitle includes the lag order and the prediction errors. In black is the case number input, blue is the ground truth and red is the prediction of the model.
• Figure 5: Sample models prediction for MM task. The $x$-axis represents time in weeks, and $y$-axis is the normalized cases number. The suptitle includes the model name and the prediction errors (MSE and MAE). In black is the case number input, blue is the ground truth and red is the prediction of the model.