PREDICTION PERFORMANCE OF RELATIONSHIPS AMONG MACROECONOMIC VARİABLES USING DEEP LEARNING AND DIFFERENTIAL EQUATIONS: THE CASE OF TURKEY

This study investigates the dynamic interactions and forecasting performance of key economic variables in Turkey from 2014 to 2024 using both linear ordinary differential equation (ODE) model and artificial neural networks (ANN). The ODE model captures instantaneous changes over time, while the ANN model accounts for nonlinear relationships among variables such as the US Dollar exchange rate, inflation, interest rates, economic growth, and geopolitical risks. Parameter estimation for the ODE system was performed using Runge-Kutta numerical solutions combined with least squares fitting, and the ANN model employed optimized activation functions for improved prediction accuracy. Comparative analysis shows that both models effectively predict trends, but the ANN model achieves better accuracy across all evaluation metrics. Impact analysis based on the ANN model reveals that economic growth and financial market indicators have strong positive effects on the exchange rate, while political uncertainties and external shocks tend to negatively impact it. These results highlight the sensitivity of the Turkish economy to both domestic and global factors, emphasizing the importance of stable economic policies to manage exchange rate volatility. The findings demonstrate that combining traditional mathematical modeling with machine learning techniques offers valuable insights and enhanced forecasting ability for emerging markets like Turkey, which face rapid structural changes and external risks.