Cardiovascular diseases (CVDs) are a leading cause of mortality worldwide, and early detection is crucial for effective treatment and management. Electrocardiogram (ECG) signals have been widely used for CVD diagnosis due to their non-invasiveness and ease of acquisition. However, the complexity of ECG signals and the presence of multiple CVDs in a single patient make the classification task challenging. Therefore, the early detection and diagnosis of cardiovascular diseases (CVDs) using electrocardiogram (ECG) signals is critical to preventing associated complications. This article proposes a novel approach for the reduced lead ECG multi-label classification task using 2D SEResnet with an attention mechanism. The proposed approach employs a 2D CNN architecture that incorporates wide SE residual networks (SEResnet) and a self-attention mechanism to extract relevant features from ECG signals with fewer leads. By evaluating a dataset with over 88,000 ECG records, the model showed high accuracy and an F1 score for identifying various abnormalities. Also, the model includes certain demographic features, which further improve the generality of the model. Overall, the model achieved an accuracy of 42.9% and an F1-score of 0.51 for a single-lead configuration and 48% and an F1-score of 0.55 for a two-lead configuration. These results indicate that the proposed approach can effectively perform with limited ECG leads and some demographic data, which could be helpful in real-world scenarios.