Informal Economy Dynamics in Central Africa and the Mediterranean: A Machine Learning Approach

The informal economy remains a stable and multidimensional phenomenon, especially in developing regions, where its dynamics are determined by a combination of institutional constraints and external shocks. The aim of the study is to develop a predictive model for assessing informal economy factors in Central African and Mediterranean countries based on machine learning methods. The methodological basis of the research includes the use of modern machine learning algorithms such as Random Forest, XGBoost, Support Vector Regression (SVR) and Elastic Net, using nested cross-validation (5-fold) and Bayesian hyperparameter optimization. The empirical base consisted of panel data for 28 countries (12 Central African countries and 16 Mediterranean countries) for the period 2005-2023. The share of employment in the informal sector was used as a dependent variable, while institutional indicators (quality of regulation, social spending, education) and external determinants (foreign direct investment, remittances, trade openness, and geopolitical risk) were used as factors. An analysis of the importance of the attributes shows that the quality of institutions and the coverage of social protection are the dominant internal predictors, while trade volatility and the influx of remittances act as critical external variables. Random Forest (R2 = 0.983; MAPE = 2.57%) and SVR (R2 = 0.982; MAPE = 2.17%) also confirmed the high accuracy of forecasting. It was found that among the factors, the geopolitical risk index has the greatest influence (up to 0.86 in correlation), as well as institutional indicators the quality of regulation (up to -0.96) and social spending (up to -0.93). The results show that external shocks can have a comparable or stronger impact on the level of informality compared to internal institutional factors.