A Machine Learning-Based Risk Classification Model for Hot Rolled Coil (HRC) Import Delays Using Supply Chain Visibility Indicators

Abstract

Delays in importing Hot Rolled Coil (HRC) raw material can disrupt production continuity and generate operational losses for manufacturing companies that depend on imported supply. This study aims to develop a risk classification model for HRC raw material import delays based on Supply Chain Visibility indicators, using a risk scoring approach combined with machine learning at PT Berjaya Mandiri Indonesia. A quantitative descriptive approach was applied to 200 historical HRC import records from 2020 to 2025. Six risk indicators were weighted through a risk scoring method to produce Low, Medium, and High-Risk labels, which subsequently served as the learning target for three classification algorithms Naïve Bayes, Decision Tree, and Random Forest evaluated using stratified 5-fold cross-validation. Decision Tree and Random Forest achieved the highest average accuracy at 97.0%, with Decision Tree marginally outperforming Random Forest on macro-F1 (0.965 versus 0.962) and consequently selected as the final model on grounds of computational efficiency and interpretability. A confirmation interview with four company practitioners indicated that the model possesses good face validity. The integration of risk scoring and machine learning produced a risk classification model that consistently automates the manual risk-scoring rules; however, the high accuracy obtained reflects consistency in replicating a deterministic label rather than predictive capability over an independent outcome, so further predictive validation is still required before the model is applied operationally.

Keywords

Supply Chain Visibility; Risk Scoring; Machine Learning; Import Delay; Supply Chain Risk Management