Electrocardiogram (ECG) classification is important for the early detection and treatment of arrhythmias and significantly contributes to cardiovascular disease detection. In this study, present a comparative analysis of arrhythmias from ECG signals utilizing four distinct deep learning and machine learning classifiers: K-Nearest Neighbors, Convolutional Neural Networks, Support Vector Machine, and Recurrent Neural Networks. The dataset utilized, MIT-BIH Arrhythmia Data-base, which includes annotated ECG recordings with five classes of beat sub-types N: Normal, S: Supraventricular ectopic, V: Ventricular ectopic, F: Fusion, and Q: Unclassifiable. Our experimental results demonstrate that DL models CNN with accuracy of 99.00% and RNN with 98.00%, significantly outperform basic machine learning approach KNN and SVM attaining an accuracy 96.00% in classification accuracy and robustness. The CNN model exhibited superior performance because of its capability to capture spatial structure in the ECG signals. Conversely, the RNN model effectively leveraged temporal dependencies, showcasing its strength in sequential data processing.