<p>Flyrock is one of the most serious consequences of surface mining and poses significant safety and operational risks. Traditional empirical models fail to capture the non-linear interactions between blasting parameters and rock mass properties, and most machine learning approaches only do forward prediction without providing design guidance. This paper introduces a new 2-phase hybrid framework that combines metaheuristic optimization with adaptive neuro-fuzzy inference system (ANFIS) modeling for both prediction and inverse blasting design. In the first phase, five state-of-the-art optimizers, including social spider optimization (SSpiderO), bird swarm algorithm (BSA), dragonfly optimization (DO), elephant herding optimization (EHO), and fireworks algorithm (FA), were used to tune ANFIS parameters. The results showed that the SSpiderO-ANFIS model was the most accurate and robust predictor of flyrock distance with accuracy of ~ 82%; the other models only had 66–80% accuracy. In the second phase, salp swarm optimization (SSO) was applied to the trained SSpiderO-ANFIS model as a surrogate fitness function to inverse optimize blasting parameters under safety constraints. The results showed that the proposed framework not only outperformed traditional empirical and machine learning models in prediction but also provided optimized blasting configurations that restricted flyrock within regulatory limits. Sensitivity analysis also showed the relative effect of burden, spacing, stemming length and powder factor on flyrock behavior. The study demonstrates the practicality of combining prediction and optimization for safe and cost effective blast design and suggests future research on multi-objective optimization and transfer learning for broader applicability.</p>