Authors: Syed Raza (UAH), Nikolai Pogorelov (UAH), Talwinder Singh (UAH)

In this study, we aim to improve the accuracy of arrival time predictions for MHD coronal mass ejections (CMEs) by using data from multiple viewpoints of STEREO coronagraph and heliospheric imager (HI). We use the kinematics derived from coronagraph observations and the graduated cylindrical shell (GCS) model to simulate a flux-rope-based CME for the 12 July 2012 event. Comparing the simulated CME with observations on Earth, we found a delay of approximately 2.5 hours in the arrival time of the simulated CME. To further compare the simulated CME with observations in the inner heliosphere, synthetic J-maps were created from simulation data and compared with J-maps created from STEREO HI observations. On average, the simulated CME was trailing the observed CME by 2 hours in both STEREO-A and B. By adjusting the arrival time of the simulated CME accordingly, the arrival time error was reduced to just 0.5 hours, demonstrating the potential of using HI observations to improve arrival time predictions in MHD models. Moreover, two machine learning techniques, Lasso regression, and Neural Networks were used, along with combinations of STEREO-A and STEREO-B data, to further enhance the accuracy of CME arrival time predictions. The mean Average Error (MAE) for the arrival time was reduced to 4.1 hours in the best case, compared to over 10+ hours using previous models. Using Neural Networks with STEREO-A data only reduced the MAE to 5.3 hours. We plan to validate this approach by performing simulations of multiple other CMEs in future work.