The significance of automated electrocardiogram (ECG) analysis technology in early prevention and diagnosis of cardiovascular diseases is increasingly evident. Deep learning algorithms have made significant strides in arrhythmia detection, enabling automatic feature extraction from raw ECG signals and precise classification. However, improving identification accuracy remains a challenge. This study proposes a CNN-LSTM model with an attention mechanism, effectively integrating convolutional layers, Long Short-Term Memory (LSTM) layers, and attention mechanisms to exploit spatial and temporal features of ECG signals, thereby enhancing accuracy and robustness in arrhythmia classification tasks. Ten-fold cross-validation on the MIT-BIH arrhythmia database demonstrates the model’s outstanding performance, achieving an accuracy of 99.31%. These findings highlight its tremendous potential in ECG analysis compared to existing models. This achievement not only advances medical diagnostic technology but also provides new perspectives and tools for early prevention and treatment of cardiovascular diseases.