A deep neural network method for the reliable categorization of cardiac arrhythmias is presented in this research. The suggested technique automatically extracts discriminative characteristics from electrocardiogram (ECG) signals and classifies them into various arrhythmia classifications using an amalgamation of convolutional and recurrent neural networks. Using the PTB Diagnostic ECG Database for training and evaluation, the model demonstrated good levels of accuracy, sensitivity, and specificity on both test and training sets. Furthermore, a range of tests, such as adversarial attacks and noise injection, were used to evaluate the model’s robustness, showing that it could continue to function well even under difficult circumstances. The suggested approach exhibits encouraging outcomes for a reliable and accurate classification of cardiac arrhythmias.