Accurate forecasting is essential for optimizing supply chain operations, directly influencing inventory control, logistics efficiency, and procurement planning. Forecasting inaccuracies can lead to stock shortages, excess inventory, and financial losses, significantly impacting business performance. Many industries, including automotive, retail, and energy, have faced disruptions due to unreliable demand predictions, resulting in production inefficiencies and supply chain bottlenecks. This study presents a data-driven deep learning framework that improves forecasting accuracy by integrating Predictive Power Score (PPS) with correlation-based feature selection to capture both linear and non-linear dependencies, ensuring that only the most predictive variables are selected. To further enhance model performance, this approach incorporates hyperparameter tuning within a Multilayer Perceptron (MLP) classification model, refining predictive accuracy through systematic optimization. By combining PPS-driven feature selection [7], data quality improvements, and hyperparameter tuning, this research demonstrates how businesses can enhance supply chain forecasting and minimize operational inefficiencies. Furthermore, the study quantifies the Return on Investment (ROI) of improved forecasting accuracy, illustrating how enhanced predictions translate into tangible financial and operational benefits. The findings highlight the necessity of a data-centric forecasting strategy that prioritizes robust feature selection and model optimization. By transitioning from traditional heuristic-driven forecasting to scalable AI-based solutions, businesses can improve supply chain resilience, reduce operational costs, and enhance decision-making efficiency in an increasingly dynamic global market.