Abstract: OBJECTIVE To establish the prediction model for antibacterial drugs-associated fever (ADF) so as to provide guidance for reasonable use of antibacterial drugs. METHODS The patients who had adverse reactions that might be ADF and were treated in Jiaozuo People's Hospital from Jan. 1, 2019 to Dec. 31, 2024 were enrolled in the study, the clinical data were extracted and randomly divided into the training set and the test set in a 7∶3 ratio. The final predictive variables were screened out by LASSO regression and logistic regression, the prediction models were established based on 9 types of machine algorithms, the optimal models were screened out. The receiver operating characteristic (ROC) curves, decision curves for decision curve analysis (DCA), precision-recall (PR)curves and calibration curves were drawn to analyze the efficiencies of the models, and the impacts of the predictive variables on the models were interpreted through Shapley additive explanation (SHAP) method. RESULTS A total of 204 patients were enrolled in the study, 96 of whom had ADF. The percentage of eosinophilic granulocyte, percentage of monocytes, difference value of red cell counts before and after drug therapy and neutrophil counts and standard deviation of red blood cell distribution width at the highest body temperature were the characteristic variables for ADF. Categorical feature gradient boosting machine (Catboost) was verified as the optimal model. The area under the curve (AUC) was 0.846 in the test set, and the prediction result with the threshold probability ranging between 25% and 78% yielded the positive clinical benefit. SHAP analysis indicated that the characteristic variable, ranking in the order of importance, were as follows: the difference value of eosinophils percentage, difference value of monocyte percentage, standard deviation of red cell distribution width at the highest body temperature, difference value of red cell counts, neutrophils counts at the highest body temperature. CONCLUSIONS The eosinophils percentage, monocytes percentage, difference value of red cells counts before and after drug therapy and neutrophils counts and standard deviation of red cell distribution width at the highest body temperature are the characteristic variables for ADF. The Catboost prediction model that is established based on the five variables can achieve the most remarkable predictive effect and, to some extent, may assist clinicians in making reasonable treatment decisions.