Hybrid <scp>BeeHive</scp> Algorithm: Proposed Ensemble Model for Multiclass Multi‐Label Ophthalmological Eye Diseases Prediction

ABSTRACT The purpose of this study is to develop a novel and effective deep learning method, named the Hybrid BeeHive Algorithm (HBA), for the intelligent identification and classification of retinal diseases using the Ocular Disease Intelligent Recognition (ODIR) dataset. The methodology involves the introduction of the HBA, which incorporates key elements from Inception, DenseNet, and VGG architectures. This study utilizes the ODIR dataset, consisting of 5000 cases from various Chinese hospitals, to train and evaluate the proposed model. The preprocessing phase includes standardizing image resolutions, normalizing pixel values, and augmenting data to enhance the model's generalizability. The HBA integrates data from color fundus photographs to create a robust multimodal system. Backpropagation is used to optimize the model's parameters, enhancing its ability to recognize complex patterns in the data. The results demonstrate the efficacy of the HBA in classifying various retinal diseases. The performance metrics for different models (DenseNet121, InceptionV3, and VGG19) are compared, highlighting their accuracy, F1 scores, precision, recall, and specificity in diagnosing age‐related macular degeneration (AMD) and other conditions. DenseNet121 and InceptionV3 models exhibit high performance, with InceptionV3 achieving near‐perfect metrics. Traditional machine learning models like RF and DT also show commendable performance but with some trade‐offs in recall. The HBA proves to be a significant advancement in the field of automated ocular disease diagnostics. Building upon existing multimodal ensemble approaches, HBA introduces a novel hybrid fusion strategy that integrates three deep CNN backbones for improved multi‐label retinal disease prediction.