We discuss a comparison of the results obtained from various Machine Learning models used for the diagnosis of Brugada syndrome, a rare heart rhythm disease, through the use of variables extracted from electrocardiograms (ECGs), either in the form of features extracted from the characteristic QRST waves of the ECG or in the form of vectors containing the averaged beats resulting from ECG signal processing. The considered ECGs were the result of an original data collection by clinical centers participating in the BrAID project, which is aimed at developing an innovative system for accurate diagnosis of the syndrome based on automatic recognition of characteristic ECG patterns of the syndrome. The models were tested and compared using different feature configurations and combinations. An analysis phase also made it possible to identify the most significant ECG features and segments. Overall, the results of the models, evaluated by double k-fold cross validation techniques, allowed to demonstrate very good performances in the context of the considered problem and may contribute to offer a new approach for supporting Brugada syndrome diagnosis.