Objective:The study was proposed to identify echocardiographic variables related to cardiotoxicity and/or death, analyzed by machine learning. Methods: The study was initially composed of a cohort of 101 patients with breast cancer, from February 2010 to June 2011, undergoing chemotherapy treatment with anthracyclines, with clinical evaluation, electrocardiogram and complete echocardiogram, including tissue Doppler and myocardial deformation indices. The exams were performed before the first session, 3, 6, 12 and 24 months, and 58 patients completed the study. Results: The mean age of the patients was 52.49 ± 12.97 years. Thus, of the 9 patients with a 10% point drop and a LVEF lower than 50% at the last moment, there were 8 with a reduction in GLS greater than 15% between the first and subsequent moments of the study, totaling 17 patients now defined as having cardiotoxicity. Associated with the outcomes of death (12 patients) and/or cardiotoxicity at the final moment (17 patients), totaling 29 patients. The random forest classifier presented the best result of the study with 77.78% accuracy, 87.89% area under the ROC curve and 80% recall, the KNN method showed 72.22% accuracy, 62.31% area under the ROC curve and 40% recall, while the XGBClassifier presented 81% accuracy, 83% area under the ROC curve and 66.6% recall. After optimizing the echocardiographic variables demonstrated by the Shap values, through the library to explain the predictions of machine learning models, it was observed that, in addition to the reduction in the LVEF, indices such as the TAPSE, e’ wave velocity of the lateral region of the mitral valve, the myocardial performance index and MAPSE, were related to cardiotoxicity and/or death. Conclusion: This study demonstrated that echocardiographic parameters obtained by transthoracic echocardiography are predictors of cardiotoxicity and death. New studies with a larger sample of patients may confirm these findings