Glaucoma is a leading cause of irreversible blindness worldwide, often progressing asymptomatically in its early stages. Early and accurate detection is critical to prevent permanent vision loss. While deep learning has shown promise in ophthalmic diagnosis, most existing models process heterogeneous clinical and imaging features monolithically, potentially overlooking group-specific patterns. This study proposes a novel multi-branch feedforward neural network (MBFNN) equipped with an attention mechanism for automated glaucoma detection. The model processes six distinct groups of multimodal features, including clinical parameters, visual field indices, and retinal layer thicknesses from swept-source OCT, through dedicated parallel branches. An attention layer dynamically learns to weight the contributions of each branch. The model was trained and evaluated on a dataset of 132 eyes (62 glaucomatous, 70 healthy). The proposed MBFNN achieved 90.0% accuracy, 100% precision, 77.8% recall, 100% specificity, 87.5% F1-score, and 83.8% AUC-ROC. It outperformed baseline models, particularly in eliminating false positives. Attention weight analysis revealed that the total retinal thickness (TRT) branch contributed most significantly (weight ≈0.37) to the model's decision. The MBFNN provides a robust, framework for glaucoma screening, potentially reducing unnecessary referrals. Future work will integrate fundus images and involve larger, multi-centric validation.