Early detection and diagnosis of lung diseases are crucial steps toward their treatment and remediation. As the fourth most common cause of deaths around the world, respiratory ailments require attention toward their timely identification. Lung sounds recorded through electronic stethoscopes have been previously used for accurate classification and to assist in the diagnosis of lung diseases. In this work, a novel augmentation technique called sequential concatenation augmentation (SCA) is introduced. In this technique, abnormal lung sound classes are concatenated among each other to increase the data volume and subsequently enhance the dataset. This augmented dataset is then fed to deep neural networks where the combination of feature-extracting capabilities of convolutional neural network (CNN) and the focus of attention mechanisms, including channel and spatial attention, contribute to an improved classification accuracy. Using this methodology, a test accuracy of 77.87% was obtained for four-class classification. The results demonstrate the robustness of the proposed approach in handling imbalanced datasets and achieving reliable predictions. This improved accuracy will aid in early and reliable diagnosis of respiratory diseases, therefore assisting in reducing mortality rates. This architecture brings out the importance of data augmentation and advanced neural networks in the medical field, making it an important domain for future research.