<p>Conventional empirical approaches often struggle to represent the highly nonlinear and uncertain response of liquefiable soils, especially within heterogeneous geological settings such as the Indo-Gangetic Basin. This study proposes an advanced, data-driven framework for estimating the factor of safety (FS) against liquefaction through hybrid Adaptive Neuro-Fuzzy Inference System (ANFIS) models optimized using four metaheuristic algorithms: Genetic Algorithm (GA), Firefly Algorithm (FF), Particle Swarm Optimization (PSO), and Biogeography-Based Optimization (BBO). The models were trained and validated using 301 field-based Standard Penetration Test (SPT) records from the Barauni Refinery site in Bihar, India, incorporating eight input parameters, including fines content, depth, effective stress, cyclic stress ratio, and corrected SPT values. Performance evaluation employed statistical indicators (R², RMSE, MAE), advanced external validation metrics, uncertainty quantification, Taylor diagrams, and regression error characteristic (REC) curves. Among the tested variants, ANFIS-PSO achieved superior accuracy and generalization, with ANFIS-GA performing comparably. ANFIS-FF and ANFIS-BBO demonstrated moderate capability but reduced robustness. The findings highlight the potential of hybrid soft computing integrated with nature-inspired optimization for reliable, interpretable, and scalable liquefaction assessment in a complex soil environment.</p>