Capacity planning in continuous manufacturing industries faces significant challenges due to the demand uncertainty, cycle time variability, and unplanned downtime occurrences. Traditional planning methods based on the nominal equipment capacity fail to provide realistic schedules, leading to production target deviations, increased operational costs, and reduced customer satisfaction. This research develops a model combining machine learning techniques and mathematical optimization for capacity planning.

The proposed framework consists of four main components: demand forecasting using Random Forest algorithm, cycle time estimation (ideal and actual) employing Random Forest algorithm, downtime pattern analysis through moving average methodology, and bi-objective capacity optimization aimed at maximizing productivity while minimizing overtime costs. The Random Forest algorithm was selected for its superior performance in handling complex, non-linear relationships and its robustness against overfitting. For demand forecasting, historical production data spanning four years was preprocessed through standardization and feature encoding techniques. The cycle time approach distinguishes between ideal conditions (effective time/output ratio) and realistic scenarios (total shift time/output ratio), providing comprehensive insights into production capabilities.

The downtime analysis component utilizes moving average techniques to identify recurring patterns and predict failure probabilities, enabling proactive maintenance scheduling. The optimization module formulates a bi-objective mathematical model that balances normal capacity utilization maximization with overtime cost minimization, subject to demand fulfillment, capacity constraints, and resource availability limitations. The model incorporates capacity transfer capabilities between months, allowing efficient resource allocation throughout the planning horizon.

Model validation was conducted in a heating radiator manufacturing facility using four years of production data. Implementation results demonstrated the effectiveness of the proposed approach, achieving a 34% reduction in downtime, 25% increase in productivity, and 106% achievement of annual planning targets. The integrated methodology successfully bridges the gap between theoretical capacity and practical production requirements, providing manufacturers with a robust tool for strategic capacity planning. The study contributes to the manufacturing optimization literature by presenting a novel combination of ensemble learning and mathematical programming techniques specifically tailored for continuous production environments.