Space-distributed machine learning based on climate lag effect: Dynamic prediction of tuberculosis

Abstract

An in-depth grasp of the complex relationship between climate and disease is crucial for fostering the development of public health security. However, certain limitations, such as overlooking spatial heterogeneity and lag effects, persist when revealing this complex relationship. With tuberculosis (TB) as a case, a pioneering space-distributed machine learning (SDML) framework is introduced to enhance TB prediction accuracy and unveil its complex relationship with climatic factors. Results demonstrate the nonlinear, intricate relationship between TB and climate variables, emphasizing the significant lag effect of climate variables on TB. Model comparisons demonstrate that SDML has a significant improvement in prediction, particularly in lag effect identification. The determination coefficient average of SDML (0.786) surpasses that of traditional machine learning (ML, 0.719). Utilizing an interpretable ML method to identify the impact of climate variables on TB, this study reveals evident spatial heterogeneity in the response of TB to climate. The spatial heterogeneity of the effects of extreme climate on TB suggests regionalized prevention and control strategies for diverse regions. This study provides a novel perspective on comprehending the intricate relationship between TB and climate, showcasing the feasibility of artificial intelligence-assisted scientific discovery.