Forecasting monthly AIDS incidence in China via LSTM-CNN parallel fusion: a comparative study of 10 predictive models

Acquired immunodeficiency syndrome (AIDS) poses a significant global public health threat and ranks among the most fatal infectious diseases in China. Effective prevention hinges upon early surveillance and predictive warning systems. However, effective predictive tools specifically tailored to AIDS incidence in China remain scarce. This study aimed to systematically compare the applicability of multiple predictive models for forecasting monthly AIDS incidence in China, identify the relatively better-performing model within this dataset, and provide preliminary methodological references for AIDS surveillance.

We collected monthly AIDS incidence data from China spanning January 2001 to September 2025. We developed and validated 10 predictive models across four categories: (I) traditional statistical approaches [seasonal autoregressive integrated moving average (SARIMA)]; (II) traditional machine learning algorithms [support vector regression (SVR) and Random Forest]; (III) deep learning models [gated recurrent unit (GRU), convolutional neural network (CNN), and long short-term memory (LSTM)]; and (IV) hybrid deep learning fusion models [GRU-LSTM serial fusion, LSTM-CNN parallel fusion, LSTM-GRU-Attention, and LSTM-CNN-Attention]. Data were partitioned into training and validation sets using a 7:3 split ratio. Model performance was assessed using the coefficient of determination (R ²), mean absolute error (MAE), and mean absolute percentage error (MAPE).

Monthly AIDS incidence in China displayed marked seasonal patterns, peaking during winter months. All models captured the general temporal trend. The LSTM-CNN parallel fusion model demonstrated relatively superior generalization performance on the validation set. Error distribution analysis further confirmed that this model achieved optimal concentration, stability, and precision in its predictions. Projections from this optimal model indicate that AIDS incidence in China will follow an upward trajectory from 2026 to 2030, followed by a deceleration in growth rate; however, incidence will remain elevated.

Among the 10 algorithms evaluated, the LSTM-CNN parallel fusion model exhibited relatively superior validation performance for forecasting monthly AIDS incidence in China, suggesting that hybrid deep learning architectures may offer certain advantages in capturing nonlinear dynamic characteristics within this specific dataset. Projections from this model indicate a slowly rising trend with fluctuations over the next decade.