Resistor Demand Prediction Using Manual Gradient Boosting Implementation for Inventory Optimization

This study proposes and evaluates a manual implementation of the Gradient Boosting algorithm for monthly resistor demand forecasting to optimize inventory in electronic component stores. Efficient inventory management is critical in a competitive market to ensure product availability and minimize costs. While conventional forecasting methods often struggle with demand instability in diverse resistor SKUs, Gradient Boosting offers robust capabilities for handling complex, non-linear patterns. Our methodology involves training the model on a small, simulated historical dataset (5 unique resistor IDs), using past demand data as features. The model's performance is evaluated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), which consistently decreased during 20 training iterations (final MAE 0.2529, RMSE 0.2938). The model successfully predicted demand for Month-4 and Month-5 using a sliding window strategy. These predictions have significant implications for reducing overstocking costs, preventing understocking, and optimizing purchasing decisions. However, the use of a very small simulated dataset is a major limitation, leading to overfitting and limiting the model's generalizability for real-world applications. This study primarily serves as a methodological illustration of the core principles of Gradient Boosting. Future work should focus on larger, real-world datasets and leveraging optimized libraries for enhanced accuracy and practical reliability