Multi-criteria in decision making is widely used in various fields, including sup-plier selection in business companies. Multi-criteria in decision-making models for supplier selection are usually built on a set of criteria, where qualitative criteria are often quantified through scoring. These decision models calculate the total scores of the evaluation objects based on the criteria weights, providing decision makers with a solid foundation for their choices. In multi-criteria decision making models such as AHP or TOPSIS, expert opinions are used to weight the criteria. This makes it difficult to automate the selection process and limits human participation in ranking the evaluation objects. This study proposes to evaluate the criteria weights based on feature importance assessment methods from classification machine learning models. Methods like LIME, Permutation, SHAP, and Gini are used to calculate total weights for the features in the process of ranking them by importance. This approach will provide a clear and scientific explanation for ranking the evaluation objects. The research utilizes 362 evaluation objectives for supplier selection in the garment & textile industry, which have been empirically tested and can generate a shortlist of potential suppliers based on previously surveyed composite criteria. This will give businesses a scientific basis for their selection decisions.