Integration of Machine Learning Models and Centralized Warehousing Strategy in Multichannel Book Distribution Optimization

Abstrak

This study aims to optimize multichannel book distribution efficiency through the integration of machine learning–based demand forecasting and centralized warehouse strategy at PT Mizan Media Utama. Using three years of multichannel sales data from offline stores, marketplaces, resellers, and events, the research employs the XGBoost algorithm to predict monthly demand for selected book SKUs. The results demonstrate that XGBoost consistently outperforms conventional forecasting methods, achieving lower Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and higher R² values, indicating improved accuracy and reliability. Comparative analysis between actual sales in 2025 and forecasted results shows that XGBoost reduces average forecast error by 20–30% compared to traditional projection methods. These accurate predictions support more effective stock allocation within the centralized warehouse, minimizing overstock and stockout risks across sales channels. The findings confirm that integrating predictive analytics into distribution planning enhances operational efficiency, improves inventory control, and strengthens data-driven decision-making. This study contributes both theoretically and practically by demonstrating how machine learning can transform conventional supply chain management into a digitally integrated, responsive, and efficient system suited for the publishing and book distribution industry.