Generated Tabular Data with Multi-GANs for Arrhythmia Classification Based on DNN Models

Globally, cardiovascular diseases (CVD) cause sudden cardiac death. According to the WHO, cardiovascular disease killed more than 32% of the world’s population in 2019. Arrhythmia is one of the symptoms of CVD that can be identified as a pattern of abnormal electrical impulses to the myocardium that can cause the heart to beat too quickly or too slowly, reducing its ability to pump blood properly and increasing mortality risk. In particular, AI, which includes Deep Learning (DL) or Machine Learning (ML), is contributing to the real-time expansion and complexity of big data in the healthcare industry. AI will make prediction easier for doctors to diagnose the symptoms of arrhythmia based on its signals. In this research, Multi-GANs give the best-generated data in 3000 epochs with the minimum percentage of loss discriminator system to identify if the data is fake or not, and also with the maximum error in loss generator system to create more variance in the generated data, including noise feature. To train the data, we use DNN models i.e. CNN, Resnet-101, and Densenet-121, which produce substantial training and validating outcomes. By testing the purpose of DNN’s accuracy, precision, recall, and Fl-score, it brings the best results with correct prediction of classes in normal, atrial premature, premature ventricular contraction, a fusion of ventricular and normal, and a fusion of paced and normal, respectively. For 20 epochs and 100 batches, we get an accuracy of 99.7609 percent for the CNN model and also 100 percent in precision, recall, and Fl-score, which is the best DNN model so far in this paper.