A comparative analysis of classical and machine learning methods for forecasting TB/HIV co-infection

Area of Science:

• Epidemiology
• Public Health
• Computational Biology

Background:

• Tuberculosis/HIV coinfection presents a significant global health challenge.
• Accurate forecasting is vital for resource allocation and intervention strategies.
• Predicting TB/HIV trends requires advanced analytical approaches.

Purpose of the Study:

• To compare the predictive performance of classical statistical and machine learning models for TB/HIV coinfection.
• To evaluate various time series forecasting models, including deep learning architectures.
• To identify the most effective models for accurate TB/HIV trend prediction.

Main Methods:

• Time series analysis using exponential smoothing and ARIMA for baseline.
• Application of machine learning models: SVR, XGBoost, LSTM, CNN, GRU, CNN-GRU, CNN-LSTM.
• Performance evaluation using MSE, MAE, sMAPE, and the Diebold-Mariano test.

Main Results:

• Deep learning models, specifically Bidirectional LSTM and CNN-LSTM, demonstrated superior performance.
• These advanced models significantly outperformed classical statistical methods.
• Accurate forecasting of TB/HIV coinfection trends was achieved.

Conclusions:

• Deep learning models are highly effective for modeling complex TB/HIV coinfection dynamics.
• Bidirectional LSTM and CNN-LSTM offer enhanced accuracy for time series forecasting in public health.
• The findings support the adoption of deep learning for improved TB/HIV surveillance and planning.