Estimates of gestational syphilis in Brazil using mathematical and artificial intelligence models
DOI:
https://doi.org/10.12662/2317-3076jhbs.v14i1.6617.pe6617.2026Keywords:
bacterial infections, sexually transmitted diseases, prediction algorithms, health planningAbstract
Objective: tto estimate new cases of syphilis among pregnant women in Brazil between July 2024 and June 2027 using Seasonal Autoregressive Integrated Moving Average (SARIMA), Long Short-Term Memory (LSTM), and Complex Exponential Smoothing (CES) models. Methods: data from the Notifiable Diseases Information System on monthly notifications between January 2014 and June 2024 were used. Validation was performed using Mean Absolute
Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) metrics. Implementation was carried out in Python. Results: the time series was found to be non-stationary (Augmented Dickey-Fuller test), exhibited an upward trend (Mann-Kendall test), and showed autocorrelation (Ljung-Box test). SARIMA (RMSE = 987; MAPE = 11%; MAE = 726) and CES (RMSE = 1225; MAPE = 14%; MAE = 875) performed well regarding error metrics, proving suitable for the series in question. However, LSTM (RMSE = 811; MAPE = 9%; MAE = 610) outperformed both. The models diverge in their forecasts through May 2027: SARIMA predicts a decline in cases, while CES predicts an increase; LSTM estimates a decrease followed by a rise in notifications. Conclusions: all tested models proved reliable for forecasting syphilis cases among pregnant women in Brazil over the next three years, but LSTM demonstrated greater accuracy. Therefore, its use is recommended to support decision-making by public administrators.
Downloads
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Maria Clara da Silva Maia, Antonio Manoel Ferreira Raymundo, Daiane Conceição de Araújo, Kelly de Almeida Schlager, Ketlin Angelin, Lídhia Cainnã de Souza Araújo, Daniela Teixeira Borges, Renata dos Santos Rabello

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.














