Automatic Purchase Order Classification Using SVM in POS System at Skus Mart

Abstract

In retail business processes, decision-making regarding Purchase Order PO submissions often remains manual and subjective, creating risks that impede procurement efficiency. The study develops an automatic classification model to predict PO approval status using Support Vector Machine SVM algorithm integrated within Point of Sale POS systems. Historical purchase transaction data was obtained from SKUS Mart POS database containing 133 entries, including attributes such as item quantity, purchase price, previous stock levels, and total purchase amounts. The research applies CRISP-DM methodology, encompassing business understanding, data exploration, preprocessing normalization using StandardScaler, model training, evaluation, and deployment phases. The model was trained using linear kernel and validated through holdout technique with 80:20 ratio for training and testing. Test results demonstrate that the SVM model achieves 76.69% accuracy, 82.21% precision, 76.69% recall, and 78.51% F1-score. The model was implemented in a web-based POS system CodeIgniter 3 combined with Python scripts to generate automatic classifications displayed directly in the user interface. Although the model demonstrates adequate performance, the study has not compared its effectiveness against other machine learning algorithms such as Random Forest or K-Nearest Neighbor. These findings establish initial groundwork for machine learning integration to support decision automation in procurement systems.

Keywords

Support Vector Machine; Purchase Order; Point of Sale; Automatic Classification; CRISP-DM