Enhancing the computational efficiency of traffic forecasting is important for urban planning and congestion management. Traditional methods often face challenges in handling the complex variable nature of the traffic flow. A novel approach has been introduced to enhance the traffic prediction method by executing and comparing the node generation methods such as random walk, DeepWalk, and Node2Vec using performance metrics. Nodes are converted into time series data and are used to construct adjacency matrices, which are processed by a GraphConvolutionalNetwork (GCN) to acquire any form of spatial features. Long ShortTerm Memory network then models temporal dependencies. We identify Deepwalk as the most effective technique. As a novel contribution, we apply Deepwalk to an existing Spatio-Temporal Graph Ordinary Differential Equation model. We compare our method with baseline and best in class algorithms by (MAE) Mean Absolute Error and (RMSE) Root Mean Squared Error and realized that our offered approach is performing better than existing approaches.