Title : Comparative analysis of time-series models for predicting dengue outbreaks in Brazil
Abstract:
Dengue remains a major public health challenge in tropical and subtropical regions, with Brazil accounting for over 70% of reported cases in the Americas during the past decade. In recent years, the number of cases has risen sharply, largely driven by the reemergence of DENV-3 genotype III after 15 years of absence. This resurgence underscores the complexity of dengue dynamics, which involve virological, environmental, and social factors, and highlights its impact on both mortality and economic costs. According to the World Health Organization (WHO), dengue is among the most persistent mosquito-borne diseases worldwide, demanding improved forecasting tools to support public health decision-making. In this study, we investigate the potential of time-series forecasting models to predict dengue outbreaks in Brazil. Historical epidemiological data were collected from the Notifiable Diseases Information System (SINAN) and the InfoDengue platform, focusing on the municipality of São José dos Campos, São Paulo, over the period from 2014 to 2024. An exploratory data analysis was first conducted to identify temporal patterns, seasonality, and recent increases in transmission intensity. Subsequently, a comparative evaluation of different forecasting methods was carried out, including statistical approaches such as Autoregressive Integrated Moving Average (ARIMA) and machine learning techniques such as Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models. The models were applied to generate forecasts of dengue incidence for 2025. By comparing their performance, the study reveals both the strengths and limitations of each approach, illustrating how artificial intelligence and Big Data can enhance disease surveillance. Such findings may help public health authorities design timely interventions and optimize resource allocation to mitigate future dengue outbreaks.
