The invention relates to the technical field of medical
artificial intelligence, and discloses an
artificial intelligence-based
voriconazole hepatotoxicity occurrence risk prediction method, which comprises the following steps of: firstly, obtaining a
medical record data set containing basic information of a patient,
medicine information, inspection information and a research end point index, and performing cleaning and conversion treatment to obtain a
medical record data set; a
random forest, a naive Bayes
algorithm and a limit gradient lifting
algorithm are respectively adopted to construct a model, the performance of the model is evaluated through 5-fold
cross validation in combination with multiple indexes, and an optimal
liver injury endpoint index and a corresponding prediction model are determined; identifying a core
risk factor by using the feature importance and an SHAP method; and finally, inputting
medical record data of a patient to take
medicine to obtain a
risk assessment result, and automatically generating a
medicine adjustment suggestion in combination with a
risk threshold. According to the method,
voriconazole hepatotoxicity risk accurate prediction and medication guidance are realized, and medication safety is improved.