Lead-free cesium bismuth iodide (Cs3Bi2I9) perovskite solar cells (PSCs) offer an environmentally friendly alternative to lead-based devices but suffer from limited power conversion efficiency (PCE) due to charge transport inefficiencies and recombination losses. This study presents a hybrid approach combining SCAPS-1D simulations and machine learning (ML) to optimize FTO/ETL/Cs3Bi2I9/GO/Au PSC architectures. Various electron transport layers (ETLs) (TiO2, ZnO, SnO2, IGZO) and graphene oxide (GO)-based hole transport layers (HTLs) were evaluated, identifying SnO2 as the optimal ETL. The optimized device achieved VOC = 1.43 V, JSC = 12.78 mA/cm2, FF = 80.49%, and PCE = 14.67%, with quantum efficiency (QE) near 100% across the visible spectrum. Performance improvements were linked to low series resistance, high shunt resistance, and effective interface engineering. ML models (Random Forest, Gradient Boosting, K-Nearest Neighbors, and Artificial Neural Network) trained on 625 SCAPS-generated datasets highlighted Gradient Boosting (GB) as the most accurate (R2 = 0.9985, RMSE = 0.038), capturing complex nonlinear interactions among device parameters such as absorber thickness, Rs, Rsh, and temperature. The integrated SCAPS-ML framework enables rapid, cost-effective optimization of PSC architectures, providing insights for defect management, transport layer selection, and stability enhancement. These results demonstrate a practical pathway to designing efficient, stable, and environmentally sustainable Cs3Bi2I9 PSCs through combined numerical and data-driven optimization.