A novel hybridized feature selection strategy for the effective prediction of glaucoma in retinal fundus images

Feature selection (FS) is crucial to transforming high-dimensional data into low-dimensional data. The FS approach selects influential traits and ignores the rest. This approach improves machine learning (ML) classifiers by reducing computational complexity and solution time. This empirical study presents a novel and effective methodology that uses two contemporary state-of-the-art soft-computing algorithms, the Grey Wolf Optimizer (GWO) and the Whale Optimization Algorithm (WOA). We have also created the hybrid version (hGWWO) of these two approaches as our novel, innovative scientific contribution. The baseline algorithms above have been used previously for feature selection across different domains. According to our understanding, these three algorithms are being used for the first time in glaucoma identification, particularly on the publicly available benchmark dataset, ORIGA. The rising global prevalence of glaucoma prompted this proposed methodology's focus on the illness. This illness is second only to cataracts in causing visual loss. Medical imaging professionals are examining retinal scans to diagnose glaucoma. Manual eye screening and retinal fundus imaging for confirmation of this infection require skilled ophthalmologists. The screening analysis method is time-consuming, requires experienced staff, and is subject to observational differences. In order to overcome these issues and to support the medical fraternity, an artificial intelligence-supported computer-aided clinical decision support system (CA-CDSS) is implemented in the present endeavor for confirmation of this disease from retinal fundus images. Nature-inspired computing strategies for feature selection and ML models for classification are employed to classify fundus retinal images under investigation. From the ORIGA dataset, sixty-five features were retrieved. A subset of most influential features is selected from the original dataset using three soft-computing-based FS methods. ML classifiers are trained using this portion of data and evaluated using a 70:30 technique. The suggested method yielded 96.8% accuracy, 0.981 specificity, 0.992 sensitivity, 0.969 precision, and a 0.982 F1-score. This study shows fresh initiatives with positive effects on ophthalmologists, researchers, and the public.