In response to the limitations of the traditional minimum data set (MDS) screening method for cultivated soil quality assessment indicators in the regional scale application, this study focused on the cultivated surface soil (0-20 cm) in the black soil area of northeastern China, and proposed a MDS screening method for cultivated soil quality assessment applicable to the regional scale based on the random forest and Shapley additive explanations (SHAP) models. The results indicated that the MDS screened by the traditional principal component analysis (PCA) method included organic matter, pH, sand, available phosphorus, and available potassium, whereas MDS screened by the SHAP method included organic matter, pH, clay, available phosphorus, and bulk density. The comparative results showed that the soil quality index (SQI) derived from the MDS constructed using the SHAP method exhibited a stronger correlation with the SQI based on the total dataset (R²=0.82). In contrast, the correlation for the SQI derived from the PCA-based MDS was only R²=0.63. This indicates that, compared to the traditional PCA method, the SHAP method can retain more comprehensive information while effectively reducing the number of soil indicators. Furthermore, SQI obtained by SHAP method showed a stronger correlation with crop yield (R²=0.40), significantly higher than the result from PCA method (R²=0.15). This indicated that MDS established by the SHAP method could more accurately reflect the actual soil quality conditions in the study area. In conclusion, MDS and SQI models constructed in this study based on SHAP method are not only practical and robust but also can provide a powerful tool for soil quality assessment and management of cultivated land in the black soil region of Northeast China.