Explainable AI for economic performance prediction evidence from industrial production, institutional quality, and consumer demand

Abstract The GWO-XGBoost model achieved an R² of 0.991 in Gross Domestic Product (GDP) prediction, exceeding all other machine learning models compared in this study. Thus, GWO-XGBoost provides a transparent and reliable decision-making support system for all those who develop economic policies. Developing accurate macroeconomic predictions can assist in the advancement of sustainable development. However, the non-linear relationships among many macroeconomic variables, such as industrial production, governance, and demand, make it difficult to develop accurate and understandable predictions using conventional macroeconomic models. To address the limitations of these traditional models, we propose a hybrid explanatory model, GWO-XGBoost, that integrates the Grey Wolf Optimizer (GWO) with Extreme Gradient Boosting (XGBoost). This integrated model automatically tunes its hyperparameters. Monthly US data from 1996 to 2020 was utilized in the experiment. These data include: Government Effectiveness, Consumer Demand/Retail Sales, Industrial Production, Trade Balance, and Unemployment. The results of the SHapley Additive exPlanations (SHAP) analysis show that the two most significant factors influencing economic performance were consumer demand and government effectiveness. As a result of being both highly predictive and having clear feature attributions, the proposed hybrid explanatory model will be useful for all institutions involved in developing and implementing evidence-based planning processes that align with economic growth.