The invention relates to the field of
artificial intelligence in
environmental health, discloses an innovative method of fusing
machine learning,
causal inference and high-dimensional interaction analysis, and discloses a high-dimensional interaction recognition-based heavy
metal mixed
exposure membranous nephropathy prediction method and a high-dimensional interaction recognition-based heavy
metal mixed
exposure membranous nephropathy prediction system. The method comprises the following steps: firstly, collecting a
data set containing multiple heavy
metal exposure data, individual information and an MN diagnosis result, and preprocessing the
data set; then, constructing a prediction model by adopting multiple
machine learning algorithms, comprehensively screening an optimal model through
cross validation and multiple performance indexes, and preliminarily evaluating feature importance by utilizing an SHAP value; furthermore, a causal
structure based on a
directed acyclic graph is introduced,
hybrid variables are controlled by applying a dual
machine learning method, and the causal effect of individualized metal exposure on MN is estimated. On the basis of the optimal model, an SHAP interaction value and high-dimensional
model representation method is applied, the nonlinear collaborative or antagonistic interaction effect between metals is systematically identified and quantified, and the pathogenic danger threshold and high-
risk exposure window of the metals are identified.