Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multitask LSTM for Arboviral Outbreak Prediction Using Public Health Data

Lucas R. C. Farias, Talita P. Silva, Pedro H. M. Araujo

TL;DR

Arboviral outbreak forecasting for dengue, chikungunya, and Zika in Recife is addressed under data scarcity. A multitask LSTM jointly performs outbreak detection and incidence forecasting using sliding-window time series from DataSUS/SINAN. The study finds that longer inputs improve dengue regression while intermediate lengths optimize classification, with the Simple LSTM often outperforming Bidirectional variants; results generalize reasonably to 2023 data. The work demonstrates a scalable, unified neural framework for data-limited public health surveillance that can inform timely interventions.

Abstract

This paper presents a multitask learning approach based on long-short-term memory (LSTM) networks for the joint prediction of arboviral outbreaks and case counts of dengue, chikungunya, and Zika in Recife, Brazil. Leveraging historical public health data from DataSUS (2017-2023), the proposed model concurrently performs binary classification (outbreak detection) and regression (case forecasting) tasks. A sliding window strategy was adopted to construct temporal features using varying input lengths (60, 90, and 120 days), with hyperparameter optimization carried out using Keras Tuner. Model evaluation used time series cross-validation for robustness and a held-out test from 2023 for generalization assessment. The results show that longer windows improve dengue regression accuracy, while classification performance peaked at intermediate windows, suggesting an optimal trade-off between sequence length and generalization. The multitask architecture delivers competitive performance across diseases and tasks, demonstrating the feasibility and advantages of unified modeling strategies for scalable epidemic forecasting in data-limited public health scenarios.

Multitask LSTM for Arboviral Outbreak Prediction Using Public Health Data

TL;DR

Arboviral outbreak forecasting for dengue, chikungunya, and Zika in Recife is addressed under data scarcity. A multitask LSTM jointly performs outbreak detection and incidence forecasting using sliding-window time series from DataSUS/SINAN. The study finds that longer inputs improve dengue regression while intermediate lengths optimize classification, with the Simple LSTM often outperforming Bidirectional variants; results generalize reasonably to 2023 data. The work demonstrates a scalable, unified neural framework for data-limited public health surveillance that can inform timely interventions.

Abstract

This paper presents a multitask learning approach based on long-short-term memory (LSTM) networks for the joint prediction of arboviral outbreaks and case counts of dengue, chikungunya, and Zika in Recife, Brazil. Leveraging historical public health data from DataSUS (2017-2023), the proposed model concurrently performs binary classification (outbreak detection) and regression (case forecasting) tasks. A sliding window strategy was adopted to construct temporal features using varying input lengths (60, 90, and 120 days), with hyperparameter optimization carried out using Keras Tuner. Model evaluation used time series cross-validation for robustness and a held-out test from 2023 for generalization assessment. The results show that longer windows improve dengue regression accuracy, while classification performance peaked at intermediate windows, suggesting an optimal trade-off between sequence length and generalization. The multitask architecture delivers competitive performance across diseases and tasks, demonstrating the feasibility and advantages of unified modeling strategies for scalable epidemic forecasting in data-limited public health scenarios.
Paper Structure (17 sections, 3 equations, 10 figures, 2 tables)

This paper contains 17 sections, 3 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Dataset
  4. Long Short-Term Memory Networks
  5. Incidence and Outbreak Definition
  6. Preprocessing and Feature Engineering
  7. Multitask Architecture
  8. Evaluation Metrics and Statistical Analysis
  9. Experimental Setup
  10. Model Configuration and Training
  11. Hyperparameter Optimization and Validation
  12. Forecasting Procedure
  13. Results
  14. Regression Performance
  15. Classification Performance
  16. ...and 2 more sections

Figures (10)

  • Figure 2: Workflow of the proposed multitask LSTM-based prediction framework. The pipeline begins with scaled epidemiological time series data, from which overlapping input windows (60, 90, and 120 days) are generated. The dataset is divided into training (2017–2022), validation (20% of training), and testing (2023) subsets. Model tuning is performed using Keras Tuner before training with time series cross-validation. The multitask LSTM network simultaneously outputs (i) a binary classifier for outbreak detection and (ii) a regressor for case count forecasting. Predicted values are post-processed via inverse normalization to restore interpretability.
  • Figure 4: Bootstrap-based 95% confidence intervals for the F1-score (dashed lines) and AUC-ROC (solid lines) across different input window sizes (60, 90, and 120 steps) using the Simple LSTM model. Results are presented separately for (a) dengue, (b) Zika, and (c) chikungunya. Each point represents the mean performance on the held-out test set, while vertical bars indicate the uncertainty derived from 1,000 bootstrap resamplings.
  • Figure : (a) Dengue
  • Figure : (a) Dengue
  • Figure : (a) Dengue
  • ...and 5 more figures