We present a novel approach for traffic forecasting in urban traffic\nscenarios using a combination of spectral graph analysis and deep learning. We\npredict both the low-level information (future trajectories) as well as the\nhigh-level information (road-agent behavior) from the extracted trajectory of\neach road-agent. Our formulation represents the proximity between the road\nagents using a weighted dynamic geometric graph (DGG). We use a two-stream\ngraph-LSTM network to perform traffic forecasting using these weighted DGGs.\nThe first stream predicts the spatial coordinates of road-agents, while the\nsecond stream predicts whether a road-agent is going to exhibit overspeeding,\nunderspeeding, or neutral behavior by modeling spatial interactions between\nroad-agents. Additionally, we propose a new regularization algorithm based on\nspectral clustering to reduce the error margin in long-term prediction (3-5\nseconds) and improve the accuracy of the predicted trajectories. Moreover, we\nprove a theoretical upper bound on the regularized prediction error. We\nevaluate our approach on the Argoverse, Lyft, Apolloscape, and NGSIM datasets\nand highlight the benefits over prior trajectory prediction methods. In\npractice, our approach reduces the average prediction error by approximately\n75% over prior algorithms and achieves a weighted average accuracy of 91.2% for\nbehavior prediction. Additionally, our spectral regularization improves\nlong-term prediction by up to 70%.\n