Objective To analyze the registration rate of pulmonary tuberculosis (TB) in Suzhou, Jiangsu province, from 2017 to 2024, and compare the application effects of the seasonal autoregressive integrated moving average (SARIMA) model, exponential smoothing state space model with Box-Cox transformation, autoregressive moving average errors, trend and seasonal components (TBATS) model, and extreme learning machine (ELM) model in the prediction of pulmonary TB in Suzhou, and provide evidence for the development of pulmonary TB prevention and control strategies.

Methods Based on the monthly registration rate of pulmonary TB in Suzhou from 2017 to 2023, the SARIMA model, TBATS model, and ELM model were fitted and established. The monthly registration rate of pulmonary TB from January to December 2024 were used to validate the models. The root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) were used to compare the fitting and prediction performance of the models.

Results A total of 23,623 cases of pulmonary TB were registered in Suzhou from 2017 to 2024, the annual registration rate showed a downward trend from 27.21/100 000 to 17.61/100 000. The annual registration peak of TB occurred during May - September, while the trough was in February in Suzhou. The RMSE, MAE, and MAPE predicted by the SARIMA model, TBATS model, and ELM model were 0.25%, 0.21%, 14.40%; 0.23%, 0.18%, 12.57%; and 0.12%, 0.10%, 6.58%, respectively.

Conclusion All the three models demonstrated good predictive performance in the prediction of the monthly registration rate of pulmonary TB in Suzhou, and the ELM model showed better prediction accuracy, and can be used to predict short-term trend of TB incidence in Suzhou.