This study aims to develop a hybrid data-driven framework for supplier risk assessment in automotive supply chains. It seeks to integrate predictive modeling with clustering techniques to proactively identify and categorize supplier risks, enabling enhanced operational stability and resilience. The research focuses on creating a decision-support tool that combines expert-validated risk indicators with advanced machine learning (ML) methods to address gaps in traditional risk management approaches.
A systematic methodology was used, combining Delphi-based expert validation (28 indicators from 36 initial factors) with ML. Data from 500 suppliers were preprocessed (K‑nearest neighbor (KNN) outlier removal, MinMax scaling), and feature selection was performed using PCA/LASSO/Ridge. Six regression models were optimized (Grid Search and Optuna), with support vector regression (SVR) showing best performance. Agglomerative clustering segmented suppliers based on risk profiles.
Support vector regression achieved superior predictive accuracy among tested models. Clustering revealed a significant future shift toward higher-risk supplier segments, with low-risk suppliers decreasing notably. The study confirms that increasing budget allocation does not automatically reduce risk or enhance resilience. Expert-validated indicators proved crucial for accurate risk assessment, while the hybrid framework effectively segments suppliers for targeted risk mitigation strategies.
This research contributes a novel hybrid framework integrating prediction and clustering for supplier risk management. It introduces expert-validated indicators through Delphi methodology and demonstrates their critical role in risk modeling. The study provides automotive manufacturers with a practical decision-support tool for proactive risk mitigation and resource allocation, bridging the gap between theoretical risk models and practical supply chain challenges.
