Wireless capsule endoscopy (WCE) is a transformative technology revolutionizing the diagnosis of gastrointestinal (GI) disorders by offering a less invasive alternative to traditional endoscopic procedures. This study utilizes deep learning techniques to classify pathological findings detected in WCE images from the Kvasir dataset, a well-recognized repository of endoscopic images extensively used in medical research. A total of 2,337 images are considered in this study, encompassing various GI tract abnormalities, including Angiectasia, Erosion, Erythema, Polyp, Ulcer, and Lymphangiectasia. The proposed model is a modified version of DenseNet which is outperformed when compared to other deep learning models. Three distinct convolutional neural network (CNN) architectures such as ResNet50, Xception, and InceptionV3 are considered for classification task and metrics are obtained for comparison. Among all, the proposed model has produced the highest results with 98.62% precision, 98.58% recall, 98.57% f1 score and 98.58% overall accuracy in diagnosing significant GI tract diseases from WCE images. To enhance the interpretability of the proposed model predictions, explainable techniques such as LIME, RISE and GradCAM has been integrated. These methodologies yield valuable insights into the decision-making processes of the proposed models, fostering trust and comprehension among medical practitioners.