AI-Driven Predictive Modelling of US Economic Growth Using Big Data and Explainable Machine Learning

Abstract

Forecasting the level of gross domestic product (GDP) growth accurately and interpretably is important in fiscal planning and economic policy. Linear econometric models build on nonlinear dependencies and hidden structures that are likely to be missed by traditional econometric models when applied to high-dimensional macroeconomic data. This paper explains the machine learning model to predict the quarterly U.S. GDP growth based on 41 macroeconomic indicators, founded on the available Kaggle dataset, U.S. Macroeconomic Factors and Growth. Four supervised models were trained and tested (Linear Regression, Random Forest, XGBoost, and LightGBM). Compared with the more complex ensemble models, the best-performing algorithm was Linear Regression (RMSE = 0.2005, R square = 0.9959, MAPE = 4.72%). SHapley Additive exPlanations (SHAP) were included in the assessment to guarantee the pursuit of transparency in model behaviour. The findings named federal surplus/ deficit, bond yields, and per capita GDP growth as major predictors that are macroeconomic drivers. SHAP summary and dependence plots were used to determine how these indicators affected predictions during the various economic cycles. In addition, feature separability was realised as binary classification of economic expansion versus contraction, which obtained an AUC of 1.00 across models. The research adds an interpretable and reproducible pipeline to the real-time forecasting of the economy and the long-term adaptation of explainable AI to macroeconomic modelling. The magnitude of these insights in planning in the public sector, policy formulation in the central bank, and financial system stability is directly applicable.