Identifying temporal patterns of dengue epidemiological dynamics in Brazilian states using unsupervised learning

Dengue fever represents a significant challenge for Brazilian public health, exhibiting heterogeneous behavior across different regions of the country. This study aimed to identify distinct patterns in the temporal behavior of dengue outbreaks in Brazil between 2017 and 2025. To this end, weekly time series of confirmed cases obtained from the Notifiable Diseases Information System (SINAN) were used. The time series were standardized by Z-Score and compared using the Dynamic Time Warping (DTW) metric. Subsequently, the hierarchical agglomerative clustering technique was applied using different link functions. The clustering was evaluated using the Silhouette, Davies-Bouldin, Calinski-Harabasz indices and the elbow method. The results revealed the existence of 9 epidemiological profiles among the Brazilian states, highlighting differences in the seasonality, intensity, and recurrence of outbreaks. This study contributes to the understanding of the temporal heterogeneity of dengue and can support more targeted epidemiological surveillance strategies.