Klasifikasi Compliance dan Clustering Risk Profiling Supplier pada Data Procurement

Abstract

The cost efficiency and supply chain quality of a company are strongly influenced by how well purchase orders comply with established policies and how consistently supplier performance is maintained. This study builds a classification model to predict the Compliance  status of a purchase order, while also grouping suppliers into risk profiles through Clustering. The data source used is the Procurement KPI Analysis Dataset from Kaggle, covering 777 rows of purchase order data from 5 suppliers. Prior to modelling, an investigation into missing value patterns pointed toward a Missing Not At Random (MNAR) mechanism on the Defective_Units attribute, so an imputation strategy accompanied by an indicator feature (missing flag) was applied instead of row deletion. Three classification algorithms were tested   Logistic Regression, Random Forest, and Gradient Boosting   while supplier risk grouping used the K-Means algorithm. Random Forest emerged as the best-performing model, recording an F1-Score of 0.899 and a Recall of 0.977, with Defect_Rate as the most decisive feature. On the Clustering side, two of the five suppliers fell into the Higher Risk category. These findings confirm that combining classification and Clustering, coupled with careful examination of Missing Data, produces a more complete understanding to support data-driven Procurement decision-making.