This research presents a comprehensive approach to classifying optical coherence tomography (OCT) images using advanced deep learning techniques. The primary objective is to enhance the accuracy and efficiency of detecting retinal diseases. The proposed method involves a multi-stage classification process that integrates several state-of-the-art architectures, including Inception-v3, ResNet-50, DenseNet-201, and EfficientNet-B3, to extract and learn hierarchical features from OCT images. Subsequently, an Ensemble method was applied by replacing and improving several fuzzy functions. In addition to the five ensemble methods referenced, we propose two enhanced versions with improved performance. We also advocate using the Weighted Average F1-score as one of the performance evaluation criteria. Extensive experiments demonstrate that the ensemble approach significantly outperforms individual models, achieving superior classification performance. The results indicate that the proposed method effectively distinguishes between different retinal conditions with high precision and recall, marking a substantial improvement over existing methods.