Heart rate segmentation in electrocardiogram (ECG) signals is important for the early detection of heart rhythm abnormalities. Traditional methods of ECG segmentation often rely on manual feature extraction, which is time-consuming and error-prone. This work is a proposed deep learning-based algorithm to combine convolutional neural networks (CNN) and long short-term memory (LSTM) networks to extract features from raw ECG signals into five distinct groups: normal (N), left bundle branching (L), right bundle branching (R), early atrial contraction (A), and early ventricular contraction (V). The CNN component is used to capture spatial features from ECG signals, while the LSTM component captures time-dependent. The classical model is evaluated using the MIT-BIH Arrhythmia Database, a widely accepted standard for ECG classification. Key performance metrics including accuracy, precision, recall, and F1-score are reported to evaluate the effectiveness of the proposed model in all groups. This study aims to improve classification accuracy by ensuring a balanced performance among different arrhythmia types. The results show that the proposed CNN-LSTM algorithm achieves higher classification accuracy compared to traditional methods.