 Quantum walks are a crucial component of modern quantum technologies, allowing for advanced quantum algorithmic applications and energy transfer on certain graphs. However, determining the possible advantages of quantum walks on arbitrary graphs is challenging and requires simulations. A new approach using machine learning algorithms specifically designed to learn from graphs has been developed to predict the quantum advantage based on graph features, enabling the identification of graphs that exhibit quantum advantage without performing any quantum walk or random walk simulations. The approach outperforms random guess in all evaluated cases, paving the way for automated elaboration of novel large-scale quantum circuits and high-efficiency energy transfer simulations in biophotonics and material science. This article was authored by Alexei Melnikov, Leonidy Fedichkin and Aleksandr Allodgins.