Authors - Tachanat Akarajaka, Komgrit Leksakul, Chichana Suedumrong, Nivit Charoenchai Abstract - Cardiac Arrhythmia has grouped into cardiovascular disease, and it occurs when the electrical signals that coordinate the heart's beats didn't work properly. Moreover, it can cause the body to have abnormalities ranging from mild to death. Cardiac Arrhythmia is among the leading death rate in the world. It can occur from several reasons such as sleep deprivation, eating fatty foods, and lack of exercise. From the structure of the heart system, it was found that there are 4 functional electrical pathways e.g., SA Node, AV Node, Bundle Branches, and Purkinje fiber. These electrical signals can be read by EKG. In this experiment, 10,000 images of EKG from PhysioBank ATM were used and divided into 4 classes as abnormal SA+ AV Node, abnormal Bundle Branches, abnormal Purkinje fiber, and Normal condition to train and compare the result of using two CNN models: transfer learning MobileNetV2 model and transfer learning Xception model. Then, these models were used to detect Cardiac Arrhythmia. As a result, the transfer learning MobileNetV2 model has an accuracy of 98.58%. Besides, the transfer learning Xception model coved an accuracy of 94.51%. It can be concluded that the transfer learning MobileNetV2 has higher accuracy than the transfer learning Xception at 4.34%.