Energy diversification, as a strategic necessity in fossil fuel-dependent economies like Iran, plays an important role in reducing structural dependency and achieving sustainable economic growth. This study discusses the causal relationship between the energy diversification index and economic growth in Iran over the period 1970–2023. A novel nonlinear approach based on deep neural networks is employed within a sliding window framework. The performance of three machine learning models (MLP, LSTM, and GRU) is assessed in predicting each variable (economic growth and energy diversification), first using only its own lags, and then by adding the lags of the other variable. The difference in prediction accuracy is quantified using a causality intensity index (Ft) and its significance is tested via the Wilcoxon test. Results reveal that including energy diversification as an input significantly improves the prediction of economic growth—particularly in LSTM and GRU models, where MSE dropped by approximately 46% and 39%, respectively, and Cohen’s d values reached 3.90 and 4.15, indicating very large effects. In contrast, adding economic growth to models predicting energy diversification did not improve, and sometimes worsened, model performance—evidenced by a negative Cohen’s d in the LSTM model. Dynamic analysis also showed that the strength of causality from economic growth to energy diversification varied over time, intensifying during periods of structural or policy change. Overall, the nonlinear neural network approach proved superior to traditional linear methods in capturing complex and time-varying relationships between energy diversification and economic growth. Advanced models like LSTM and GRU yielded stronger and more accurate causal insights. The findings support the "growth hypothesis," indicating that energy diversification can act as a driver of economic growth in Iran. Accordingly, policymakers are advised to foster institutional support and invest in renewable energy to promote sustainable development.

JEL Classification: C45, C22, Q43, Q56, O13.
Keywords: Energy Diversification, Economic Growth, Nonlinear Granger Causality, Machine Learning, Neural Network.