Penggunaan Algoritma Machine Learning untuk Optimalisasi Rantai Pasokan di PT. Sinar Sosro

Abstract

This study aimed to develop an integrated supply chain optimization model at PT. Sinar Sosro Gunung Slamat by utilizing a machine learning approach. The developed model encompassed three main components of the supply chain: demand forecasting, inventory management, and product distribution. In the forecasting stage, the Random Forest algorithm was used to predict demand for Teh Botol Kaca based on historical variables such as sales, price, weather, seasonal trends, per capita income, and population size. The prediction results showed a very high accuracy rate of 98.76% with an error of only 1.24%, providing a solid foundation for more precise production planning. Inventory management was performed using the Support Vector Regression model, which successfully reduced the risk of overstock and stock-out by 100%, with simulations demonstrating effectiveness in maintaining inventory balance, despite relatively high MSE and RMSE values due to the large data scale. Distribution was optimized through clustering using the K Means method and determining the shortest delivery routes with the Nearest Neighbor algorithm. As a result, all delivery routes remained below the maximum threshold of 1000 km and met the company's Service Level Agreement (SLA). The system was also adaptive to changes in demand and location conditions. The novelty of this study lay in the selection of algorithms tailored to the characteristics of the data, the integration of prediction-inventory-distribution models into a single automated system, and the system's ability to evaluate and make real-time decisions based on stock thresholds and SLA. Thus, the developed model not only improved operational efficiency but also supported data-driven strategic decision-making in managing the supply chain of the ready-to-drink beverage industry.