Abstract—A mixed Support Vector Machine (SVM)-based

hierarchical learning approach is proposed to detect abnormal ECG beats. A global bi-class support vector classifier is first trained using ECG beats from different patients in a database. Then, a local novelty detector SVM is trained using only normal ECG beats from a specific patient. The fusion of the global and local classifiers can significantly improve the classification. Preliminary experimental results using MIT/BIH ECG database demonstrated good performance of our proposed approach. Keywords—Support vector machines, ECG beat recognition, novelty detection, decision fusio