Developed a machine learning-based system using TabNet and LightGBM to optimize inventory in smart supply chains. The model predicts demand, reduces stockouts and overstocking, and enhances decision-making using ERP data. Evaluated using RMSE and R² for high accuracy and adaptability.