In the current landscape of global supply chain volatility, accurate demand forecasting has transitioned from an operational advantage to a fundamental strategic necessity for industrial sustainability. This study presents a robust Analytical Decision Support System (ADSS) specifically engineered to mitigate the systemic overestimation bias often inherent in OEM customer portal data. The proposed framework addresses the limitations of individual forecasting models by implementing a multi-scale, integrative approach across both 52-week and 12-month horizons. The methodology systematically evaluates and compares three distinct algorithmic paradigms: Random Forest for its ensemble stability, Support Vector Regression for its generalization capability, and Long Short-Term Memory networks for their ability to model long-term temporal dependencies. The core of this research involves a Hybrid Ensemble Model that harmonizes the strategic intent of customer targets with the statistical discipline of machine learning through a meta-regression architecture. Performance is rigorously validated using the Mean Absolute Percentage Error (MAPE) metric against actual shipment data. The empirical findings demonstrate that this hybrid approach significantly reduces forecasting errors, achieving a monthly MAPE of 7.13 percent and outperforming the baseline customer targets, which exhibited an error rate of 10.35 percent. These results indicate that the proposed system provides a reliable and scalable tool for decision-makers, particularly in volatile demand environments where relying solely on customer-provided targets often leads to excess inventory and increased bullwhip effects. The study demonstrates that integrating diverse algorithmic perspectives provides a high-fidelity forecasting foundation, ensuring both shipment reliability and inventory optimization for the 2026 strategic period.

Keywords

Machine Learning (ML), Demand Forecasting, Model Performance and Operations Management.