The early detection of gastrointestinal abnormalities is critical for effective patient management, particularly in conditions such as bleeding, polyps, and ulcers. This study proposes an innovative approach to multi-class abnormality detection in video capsule endoscopy by utilizing an enhanced InceptionResNetV2 model integrated with attention mechanisms. The model leverages the power of deep learning to analyze frames captured by video capsule endoscopes, focusing on ten distinct classes of abnormalities: Angioectasia, Bleeding, Erosion, Erythema, Foreign Body, Lymphangiectasia, Normal, Polyp, Ulcer, and Worms. The training dataset is enhanced using data augmentation techniques, which strengthens the model's resistance to overfitting. The proposed system is trained on a dataset provided as part of a research challenge and evaluated using various performance metrics, including confusion matrices and AUC-ROC curves. Results indicate that the attention mechanism significantly enhances the model's ability to classify abnormalities accurately, achieving a validation accuracy of 92.2%. This research advances the field of medical imaging by offering a scalable and effective solution for automated gastrointestinal abnormality detection, ultimately aiding clinicians in enhancing diagnostic precision and treatment outcomes.