| Abstrakt: |
The use of machine and deep learning models for the automated detection of eye disorders has gained significant popularity. Glaucoma, cataracts, and diabetic retinopathy are prevalent ocular conditions that can result in substantial harm. Prompt identification of ocular disorders is crucial for optimal therapeutic intervention. A novel idea of implementing weighted ensemble deep learning, which is integrated with Two-Dimensional Discrete Wavelet Transform (2D DWT) and 2D Principal Component Analysis (2D PCA) for optimum feature selection for classification, has been proposed in this research. This work includes evaluating the performance of pre-trained deep-learning models in detecting eye problems using the Efficientb6 and Densnet169 models. Then, the impact of applying weighted ensemble learning on improving these pre-trained deep-learning models' eye disease classification performance is investigated. The proposed methodology comprises three essential stages: pre-processing, feature extraction, and classification. Various methods are utilized during the pre-processing stage, including implementing Data Augmentation, which involves modifying the training data. Different techniques are employed in the feature extraction stage to extract the useful features from the input data. Each of these techniques achieved different dimensionality reduction. Finally, both models of Efficientb6 and Densnet169 are employed for the extracted features to achieve exceptional performance. The proposed system accomplished a recognition rate of 96.1 with 0% dimensionality reduction when the ensemble learning model was implemented directly on the pre-processed data set, skipping the feature extraction stage. Furthermore, the system proposed achieved an accuracy of 94.9 with 67.3% dimensionality reduction when only one level of 2D DWT decompositions is used. Moreover, the system proposed obtained a comparable accuracy of 92.9 with 93.75% data reduction when the 2-level of 2D DWT decompositions is integrated with 2D PCA. The experiments undertaken in this research successfully achieved its aims of obtaining favorable outcomes of eye disease classification, reducing the data dimensions to decrease the deep learning model trainable parameters, and implementing complexity requirements. The findings of the suggested approach surpassed those shown in prior studies utilizing the same Kaggle datasets. [ABSTRACT FROM AUTHOR] |
