Objective To develop an accurate and efficient binary classification model for predicting the binding capacity of endocrine-disrupting chemicals （EDCs） to the G protein-coupled estrogen receptor （GPER）.Methods GPER binding data for 224 compounds were collected. Based on molecular descriptors and Molecular ACCess System fingerprints （MACCS）， six machine learning algorithms including random forest （RF）， artificial neural network-back propagation （ANN-BP）， extreme gradient boosting （XGBoost）， support vector machine （SVM）， k-nearest neighbors （k-NN）， and linear discriminant analysis （LDA） were employed to construct binary prediction models.Results RF， SVM， ANN-BP， k-NN and XGBoost models built with MACCS fingerprints achieved accuracies >90% and Areas of under curve （AUC） values >92% in 10-fold cross-validation， while the RF model reached 85% accuracy on the external test set. SHapley Additive exPlanations （SHAP） analysis indicated that molecules containing at least one hydrogen-bearing oxygen atom， an 8-membered or larger ring system， and a tertiary （or higher） carbon center within the ring are favorable for GPER binding.Conclusion Based on structural representation and model performance evaluation， the RF classification model built upon MACCS fingerprints was identified as the optimal model. SHAP analysis revealed that molecules containing at least one hydrogen-bearing oxygen atom， 8-membered rings or larger cyclic structures， and tertiary carbon （or higher connectivity） centers within rings may mimic certain structural features of estrogen， thereby facilitating interaction with estrogen receptors. These findings provide a theoretical foundation for the endocrine disruption mechanisms of EDCs and offer a methodological tool and theoretical basis for the targeted screening of emerging contaminants in food.