Background and aims: Biguanides and Sulfonylureas are the most commonly prescribed drugsto treat type ۲ diabetes. Clinical manifestations of anti-hyperglycemic agents may overlap eachother therefore distinguishing exposure to these drugs may be complicated. In medical toxicology,few studies to date have used machine learning (ML) on national poisoning data to identify thepotential cause of the poisoning. ML is a subfield of artificial intelligence (AI) which is appropriateto risk modeling in field of medicine. ML aims to learn relations and patterns from availabledata and computational algorithms. In This study we used the US national database to evaluatethe effectiveness of machine learning in identifying antihyperglycemic agents. With ML, doctorscan broaden diagnostic considerations, start treatment earlier, reduce adverse events, and improveclinical outcomes.Method: In this cross-sectional descriptive-analytic study, the data of single exposure by Sulfonylureasand Biguanides during ۲۰۱۴-۲۰۱۸ (n=۶۱۸۳) was demanded from the National PoisonData System (NPDS). Four machine learning models (Random Forest classifier, k-Nearest Neighbors,Xgboost classifier, Logistic Regression) were applied for this study. The XGBoost modelapproach is built on gradient-boosted algorithms and decision trees. Random forests are treebasedalgorithms that prevent overfitting. Logistic regression algorithm predicts the likelihood ofan occurrence. K-Nearest Neighbors is a fundamental supervised learning classification method.We divided the data set into two parts: Train (۷۵%) and Test (۲۵%). The performance metrics usedwere accuracy, specificity, precision, recall, and F۱-score. The “area under ROC curve” (AUC)was also used as a performance criterion that reflects how well a studied method can diagnosethese two drugs.Results: ۳۳۳۶ biguanide and ۲۸۴۷ sulfonylurea exposure were reported in National Poison DataSystem in five years. The applied algorithms for classification results of different models to diagnoseantihyperglycemic agents were accurate (۹۱-۹۳%). The accuracy of our models in determiningmentioned two antihyperglycemic agents was ۹۱-۹۳%. The Precision-Recall curve respectivelyshowed average precision of ۰.۹۱, ۰.۹۷, ۰.۹۷, and ۰.۹۸ for k-Nearest Neighbors, logisticregression, random forest, and XGB. The logistic regression, random forest, and XGB had thehighest AUC (AUC=۰.۹۷) among both biguanides and sulfonylureas groups. We have also introduceda web application (https://aiandhealth.net/applications/Antihyperglycemics_Distinguisher)to assist physicians distinguish these two agent’s poisoning.Conclusion: Machine learning can distinguish antihyperglycemic agents, which may be beneficialfor physicians without significant background in medical toxicology. Also, our suggestedWeb application has the potential to help physicians in their diagnosis.