Reconstruction and interpretable analysis of international digital trade development level measurement based on machine learning and SHAP algorithm

The purpose of this paper is to evaluate the performance of machine learning model and put forward an interpretable prediction framework for the development of international digital trade. Based on the digital trade data of 55 countries from 2010 to 2023, the evaluation index system of international digital trade development level is constructed from seven dimensions, including digital trade potential. Through XGBoost and SHAP machine learning models, the importance of different variables to the development level of international digital trade is identified. The results show that: 1) from 2010 to 2023, the development trend of international digital trade is relatively fast, and the overall score has increased from 0.079 in 2010 to 0.122 in 2023, with an increase of 54.4%; 2) From the comprehensive score of each dimension, the score of digital trade infrastructure is always ahead and the overall score shows a steady and slight upward trend, and the score of digital trade barriers is low, which is the weak link in the development of digital trade. Based on this, in order to improve the development efficiency of international digital trade, this paper puts forward five suggestions, including strengthening the cultivation and transformation of effective patents.