<p>Isomorphism identification in epicyclic gear trains (EGTs) plays a crucial role for the structural synthesis and analysis of geared mechanisms since last two decades. Although many researchers have been suggested traditional approaches based on graph theory and matrix operations often face computational inefficiencies in handling complex gear structures. This paper introduced an automated approach for detecting isomorphism in polyhedral representations of EGTs using supervised machine learning. A curated dataset comprising various classes of conventional EGT graphs is used for training and classification through Graph Neural Networks (GNNs). To validate the classification results, the Total Distance Technique (TDT) is employed, wherein the TDT value corresponds to the sum of all elements in the adjacency matrix. Two EGTs are classified as isomorphic if both their GNN embeddings and TDT values align; otherwise, they are considered distinct. All results are fully satisfied with existing literature and the proposed approach is best applicable for optimising the design of mechanical transmission systems through machine learning-driven automation.</p>