<![CDATA[Stock management is a crucial activity in the supply chain of any company, including PT. Maktal, which operates in cement distribution. The stock management system, which still relies on experience and manual methods, has the potential to cause a mismatch between demand and supply, ultimately leading to excessive inventory costs (overstock) or product shortages (stockout). The implementation of Machine Learning offers a solution to enhance the accuracy of stock needs planning. This study aims to develop and compare the performance of machine learning models, specifically the Decision Tree (C5.0) and Random Forest algorithms, in predicting the category of cement stock needs (Low, Medium, High) based on historical transaction data. The data used are historical cement sales and ordering transactions of PT. Maktal from 2020 to 2024. The stock quantity data was converted into categorical variables (Low, Medium, High) through a discretization process. Both algorithms were tested and evaluated for their performance using accuracy, precision, recall, and F1-score metrics through a cross-validation test. The comparative results indicate that the Random Forest algorithm provides the best prediction performance with an accuracy level reaching 79.91%. This performance is significantly higher than that of the Decision Tree algorithm. Feature importance analysis identified that the Purpose (customer type) and Month variables are the most influential predictors of the stock needs category. The Random Forest model proved to be effective and reliable as a data-driven decision support system to optimize stock planning and cement purchasing at PT. Maktal, reducing the risk of losses due to demand uncertainty.]]>
