Table 4 Notations used in the Pseudocode and its corresponding description.

D = {X_i,Y_i}

The input dataset, where X_i, is a fundus image and Y_i, is the corresponding class label (one of 8 diseases).

N

Total number of samples in the dataset.

D_train, D_test

The training (80%) and testing (20%) subsets of the dataset.

M

Transfer learning models: DenseNet201, EfficientNetB3, and InceptionResNetV2.

F_i

Features extracted by model M_i

L_i

LDA-selected features (7 features from each model’s feature set).

k

Number of features selected by LDA (set to 7).

L_combined

Concatenated feature set from all three models.

L_final

Final 7 discriminative features obtained from the concatenated feature set.

C

Classifiers: Deep Neural Network (DNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM).

C_model

Trained classifier model.

EVAL_train, EVAL_test

Evaluation metrics (accuracy, precision, recall, and loss) on training and testing data.

argmax(⋅)

Selects the classifier with the highest testing accuracy.

best_model

The classifier with the best performance on the test set.