Scheduling decisions in Industry 4.0 environments are increasingly exposed to uncertainty arising from dynamic shop-floor conditions, making deterministic plans prone to performance loss during execution. This study focuses on a stochastic single-machine scheduling problem with sequence-dependent setup times and proposes a smart simheuristic that combines a genetic algorithm (GA) with simulation-based evaluation. To use simulation budgets efficiently, the method allocates replications through Optimal Computing Budget Allocation (OCBA) and stores evaluated solutions in a cumulative simulation memory, enabling incremental refinement without redundant re-simulation. In addition, a Random Forest (RF) screening layer filters the OCBA candidate pool so that simulation effort is concentrated on more promising solutions. The final schedule is re-evaluated with additional simulation replications and reported with 95% confidence intervals to support reliable decision making. Extensive experiments on instance groups of varying sizes and variability levels compare the proposed variants against baseline GA and GA+OCBA approaches. Results show that memory-assisted OCBA improves solution quality under fixed budgets, and RF screening reduces simulation replications while maintaining comparable performance. Nonparametric statistical tests confirm the significance of the observed differences across instance groups. The proposed framework provides a practical and statistically supported scheduling approach for data-driven, uncertainty-aware decision making in Industry 4.0 production systems.