Master Production Scheduling (MPS) remains a cornerstone of manufacturing planning, yet its effectiveness is often undermined by uncertainty arising from demand, supply, and production-side disruptions. Existing approaches typically address these uncertainties in isolation, limiting their robustness in dynamic industrial environments. This paper proposes a Modified Separated Estimation and Optimization (MSEO) framework that extends the SEO paradigm by explicitly incorporating production-side uncertainties such as capacity limitations, crew unavailability, machine breakdowns, and product changeovers. The framework integrates machine learning–based probabilistic forecasting with stochastic optimization in a rolling-horizon setting, enabling adaptive and resilient production schedules. By uniting forecasting and scheduling into a single decision-support system, the proposed framework provides a data-driven approach for uncertainty-aware MPS. The proposed MILP model was validated through a realistic numerical case study involving three products over five planning periods under multiple disruption conditions. The model achieved an optimal objective function value of $5398.60 with a computation time of 0.0121 seconds, demonstrating its ability to effectively adapt production decisions under disruption scenarios while minimizing profit loss. The proposed framework advances MPS research by enhancing operational resilience and scheduling robustness in modern manufacturing systems.