Forecasting economic growth in Iran's volatile economy has persistently been a challenge. Aiming to identify more effective tools, this study conducts a comparative evaluation of two distinct paradigms: traditional econometrics and modern machine learning. Using annual time-series data from 1991 to 2023, an optimized Autoregressive Distributed Lag (ARDL) model was benchmarked against two advanced algorithms, Random Forest and XGBoost. The models were evaluated on a test set (2016-2023) using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) as performance metrics. The results indicate that machine learning models exhibit superior predictive performance, with Random Forest emerging as the most accurate model. Furthermore, feature importance analysis from this model revealed the key role of variables such as government expenditure, suggesting the presence of significant non-linear relationships overlooked by the linear model. The findings underscore the complementary nature of the two approaches: econometrics for interpretation and machine learning for precise forecasting.