Data science for Trade Efficiency in Developing Economies—The Case of Rwanda

This paper explores the integration of data science methods to increase trade efficiency, with a focus on optimising workflows, studying trade dynamics, and forecasting trade volume. The growing realisation of how important trade is as an asset with emerging markets is evident in cases such as Rwanda, which today receive economic influence not only from regional borders but also globally, thus drawing even greater opportunities and certain challenges. A case study based on UN Comtrade trade data from Rwanda provided an avenue for the use of machine learning models such as Hopfield networks, restricted Boltzmann machines (RBMs) and backpropagation neural networks (BPNNs) in determining the strengths and weaknesses of data analysis and prediction on trade data. This study uses the CRISP-DM (Cross-Industry Standard Process for Data Mining) analytical framework to address trade challenges such as data understanding, data preparation, modelling and evaluation. The results demonstrate that the Hopfield network, which is excellent for solving problems of mathematical optimisation, particularly trade imbalance minimisation, has its limitations in regard to trade classification problems and predictive tasks. The restricted Boltzmann machine (RBM) achieves moderate performance in classification considering accuracy and AUC-ROC score; however, the best model proposed after the analysis is the backpropagation neural network (BPNN), which is built as the best performing model since it has extraordinarily remarkable accuracy, thus being able to predict trade efficiency with a high degree of reliability. Considering the theoretical frameworks, research gaps and case studies presented in this paper, there is sufficient evidence to support the application of data science to accelerate trade, promote actionable insights for policymakers and stimulate the debate on economic development as far as Rwanda and other developing economies are concerned.