Forecasting Morocco–Africa trade flows with hybrid predictive and generative machine learning models

Moroccan foreign direct investment (FDI) in Africa has expanded considerably over the past two decades, yet its effect on bilateral trade flows remains insufficiently understood. This study aims to quantify the impact of Moroccan FDI on trade with 27 African countries and identify the key economic and institutional determinants driving these flows. A hybrid modeling framework was employed, combining predictive models (MLP, LSTM, Random Forest, Gradient Boosting Regressor) with generative models (Variational Autoencoders and conditional GANs). Model performance was evaluated using MAE, MSE, RMSE, \ (R²\), MAPE, Explained Variance, and Median Absolute Error, while SHAP analysis provided insights into variable importance. The Random Forest–VAE hybrid achieved the highest predictive accuracy (\ (R²\) = 0. 8594, RMSE = 0. 4104). SHAP analysis identified Morocco’s outward FDI (OFDI) and African partners’ GDP as the primary drivers of trade flows, whereas bilateral investment treaties (BITs) notably influenced exports. Residual diagnostics indicated minor heteroscedasticity and low bias, confirming model reliability. These findings demonstrate that hybrid predictive-generative approaches can accurately forecast trade dynamics and provide actionable guidance for policymakers and investors aiming to optimize investment strategies and enhance Morocco–Africa economic integration.