An essential part of the digestive system, the esophagus permits food and liquids to move easily from the throat to the stomach. Traditional diagnostic techniques, such as endoscopy and biopsy, are invasive and resource-intensive, making them less practical in resource-constrained environments. To address these challenges, we proposed utilizing a hybrid Deep Learning (DL) model called GAN-EfficientNet, which combines the power of Generative Adversarial Networks (GAN) and EfficientNet to effectively classify esophageal diseases. Our approach employs advanced image preprocessing techniques and data augmentation methods to enhance the diversity and balance of the dataset. Feature extraction using the Spatial Gray Level Dependence Method (SGLDM) and Principal Component Analysis (PCA) significantly reduces computational complexity without compromising accuracy. We explored both binary and multiclass classification methods, recognizing their crucial roles in various diagnostic scenarios. Binary classification distinguishes between normal and diseased states, essential for initial screening, while multiclass classification identifies specific esophageal diseases, aiding in personalized treatment plans. The GAN-EfficientNet model demonstrated exceptional performance, achieving an impressive F1 score of 99.44% for binary classification and 99.64% for multiclass classification. Additionally, we developed a web application that facilitates quick and precise disease diagnosis, particularly advantageous in areas with limited resources. This tool significantly improves patient outcomes through early detection, leading to reduced healthcare costs and less burden from advanced esophageal diseases.