INTRODUCTION:
In recent years, Paraguay has experienced sustained and intense transmission of dengue, with prolonged epidemics and peaks exceeding historical averages. The country has reported hundreds of thousands of cases, high incidence rates, and the circulation of multiple dengue virus serotypes, reinforcing an endemic–epidemic pattern. Concurrently, major chikungunya outbreaks have occurred since 2022–2023, with widespread territorial dissemination and a substantial disease burden. The co-circulation of these arboviruses, transmitted by Aedes aegypti, highlights persistent vulnerabilities in vector control and underscores the growing public health challenge posed by arboviral diseases in Paraguay.

METHODS:
This study evaluates and compares multiple predictive models applied to time series of Dengue and Chikungunya cases in Paraguay at different regional levels. Statistical models (hypothesis contrast, ARIMA), machine learning models (Random Forest, SVR), and deep learning models (ANN, LSTM) were implemented. Hyperparameter tuning was carried out using genetic algorithms. For simplicity, evaluation was mainly based on the Mean Absolute Error (MAE).

RESULTS:
The results show that hybrid models and LSTM offer the best performance in highly seasonal scenarios. In the comparison of model performance, ANOVA results reveal that the selection of a model and regional level significantly impacts the error, confirming the hypothesis that different regions and different diseases might need different models to obtain good performance.

CONCLUSIONS:
This work demonstrates the potential of computational approaches for public health management in Paraguay