The invention discloses a method for predicting
preeclampsia risk, and relates to the technical field of intelligent
medical treatment and predictive
medicine. According to the method, a three-order composite
mathematical model of static feature weighting, dynamic time attenuation and linear
probability mapping is constructed for the first time, the model is complete in structure, the three core problems of feature contribution degree quantification,
time effect correction and
risk probability nonlinear conversion are scientifically solved, and the limitation of a traditional simple
linear model is broken through; according to the method, through multi-dimensional data fusion and
machine learning optimization, the model obtains AUCgt on an independent
test set; 0.93, the accuracy is gt; the method has excellent performance of 90%, the
false positive rate is reduced to 8% or below, the method is far superior to an existing
clinical method, and a reliable basis is provided for early accurate intervention; according to the method, the dynamic time
attenuation coefficient Ktime is introduced, so that the model has the
perception ability in the time dimension, and detection results at different time points during
pregnancy can be scientifically corrected and longitudinally compared.