Deep learning classification models are extensively utilized for the automated diagnosis of heart disease (HD) by analyzing various physiological signals, such as electrocardiogram (ECG), magnetocardiography (MCG), heart sounds (HS) and impedance cardiography (ICG) signals. In this study, we introduce the ONNYX framework (Optimal Neural Networks Yielding eXplainable insights from ECG signals-based data streams), which demonstrates a big data strategy for ECG classification. This framework incorporates several modules, including FastAPI, MinIO, mlflow, Ray, Kubernetes, and Pulsar. We have developed a high throughput and low latency system using Kubernetes’ distributed architecture and Ray’s distributed training to classify ECG signals. The ECG records of subjects sourced from the MIT-BIH repository are sampled and input into the classification models to distinguish between normal and abnormal heart rate patterns in patients. We introduce an innovative optimal model selection algorithm that assesses classification techniques according to training efficiency and identifies the most suitable ones for testing. Our weighted ensemble method attained an overall accuracy of 99.27% and 99.16% in binary and multiclass classification settings respectively.