APPLICATION OF MACHINE LEARNING TECHNIQUES IN FORECASTING MACROECONOMIC INDICATORS

Abstract

Predicting macroeconomic indicators such as Gross Domestic Product (GDP), inflation, and unemployment is crucial for effective policymaking and financial planning. Traditional econometric models like ARIMA and VAR have dominated the field for decades but often struggle to capture the non-linear complexities and high-dimensional interactions inherent in modern global economies. This paper explores the transition from classical econometrics to Machine Learning (ML) methodologies, including Random Forest (RF), XGBoost, and Long Short-Term Memory (LSTM) neural networks. The study analyzes the comparative accuracy of these models, highlighting their ability to handle large datasets (Big Data) and unstructured variables. Results suggest that while ML models significantly outperform benchmarks in stable and high-volatility periods, hybrid models—combining traditional statistical rigor with ML flexibility—offer the most robust results for long-term forecasting.