Integrating predictive machine learning with transparent decision-making: A LightGBM–TOPSIS approach to supplier selection

ABSTRACT Selecting the right supplier is challenging. It is not just a matter of cost; it is a multi-criteria puzzle. In the past, traditional MCDM techniques were used for supplier selection. These techniques are streamlined and interpretable, but they are limited in today’s big data and rapidly changing markets. Machine learning, on the other hand, works very well with big, dynamic data, but it often acts like a ”black box”—it is not very transparent. Therefore, we developed a hybrid model that combines LightGBM and TOPSIS. LightGBM is a machine learning algorithm that is used for prediction, and MCDM is used for structured, interpretable decision-making. The model proposed in this article is compared in two stages. First, the machine learning component was optimized, where LightGBM outperformed six other regression algorithms (R² = 0.8823). Second, the overall hybrid model was compared with traditional MCDM methods (AHP, VIKOR, TOPSIS, BWM). The hybrid model performed well on key supplier selection criteria, achieving a hit ratio of 0.96, a Spearman correlation coefficient of 0.97, and a decision confidence of 0.94. These results confirm the effectiveness of the model for robust and data-driven supplier evaluation.