Electrocardiogram (ECG) signals play a crucial role in diagnosing various cardiovascular diseases. With the advancements in deep learning techniques, there is growing interest in utilizing these methods for automated ECG classification. In this study, we aim to provide a comprehensive analysis of ECG classification using a 4-class classification of the PhysioNet Challenge 2017 dataset, which encompasses a diverse range of cardiac arrhythmias. Our study focuses on leveraging state-of-the-art deep learning architectures, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), to extract meaningful features from raw ECG signals. We preprocess the dataset to enhance signal quality and remove noise, ensuring robust model performance. We conduct experiments involving the training and evaluation of multiple deep learning models on the PhysioNet Challenge 2017 dataset. Our approach explores different network architectures, hyperparameters, and training strategies to optimize classification accuracy and generalization performance. Additionally, we investigate the transferability of pretrained deep learning models on other ECG datasets to assess the robustness of our approach across different data distributions. Furthermore, we analyze the interpretability of our models to gain insights into the learned representations and their clinical relevance. Moreover, we address the challenge of noise and reverberation in the dataset, aiming to develop a robust system capable of performing well in real-time conditions where signal quality may be compromised. Overall, our study underscores the potential of deep learning-based ECG classification for enhancing cardiovascular healthcare in real-time conditions, facilitating more efficient diagnosis and treatment of cardiac conditions.