An Agreement-Driven Hybrid Machine Learning Framework for Glaucoma Classification Using Retinal Biomarkers
Abstract
The growing burden of glaucoma and its irreversible nature have increased the the need for trustworthy computational models that can facilitate early automated diagnosis. This work proposes an Agreement-Driven Hybrid Ensemble (ADHE) framework for glaucoma classification using pre-established biomarker features. The proposed work integrates
heterogeneous machine learning classifiers using a confidence-aware agreement mechanism combined with validation-based threshold calibration. Irrespective of conventional ensemble models that optimize a single accuracy-driven decision boundary, ADHE explicitly accounts for inter-model agreement and uncertainty to reduce clinically critical false negative predictions. Results obtained from experimental evaluation demonstrates that the proposed ADHE model has an Area Under Curve (AUC) of 0.868, an F1-score of 0.788 and an accuracy of 78.32%, while consistently reducing false negatives compared to individual classifiers and ungated fusion variants.
These results demonstrate that agreement-driven hybrid learning offers a robust, interpretable, and medically aligned solution for automatic glaucoma diagnosis in data-driven ophthalmic decision support systems.
heterogeneous machine learning classifiers using a confidence-aware agreement mechanism combined with validation-based threshold calibration. Irrespective of conventional ensemble models that optimize a single accuracy-driven decision boundary, ADHE explicitly accounts for inter-model agreement and uncertainty to reduce clinically critical false negative predictions. Results obtained from experimental evaluation demonstrates that the proposed ADHE model has an Area Under Curve (AUC) of 0.868, an F1-score of 0.788 and an accuracy of 78.32%, while consistently reducing false negatives compared to individual classifiers and ungated fusion variants.
These results demonstrate that agreement-driven hybrid learning offers a robust, interpretable, and medically aligned solution for automatic glaucoma diagnosis in data-driven ophthalmic decision support systems.
Keywords
Glaucoma classification, hybrid ensemble learning, agreement-driven decision fusion, biomarkers, machine learning