At present, the medical field is undergoing a notable shift from the traditional apprentice model to a competency-based model for surgical training. Both of the aforementioned models rely on expert surgeons for assessment. The resulting system is difficult to scale, time-consuming, and susceptible to biases, which has created the need for an automated system capable of accurately and objectively assessing surgical skill. Moreover, in supporting the surgical training process, the ability to predict future skill levels based on prior trainee assessments is invaluable for tailoring feedback and devising improvement strategies. This study has designed a recurrent neural network model to predict the skill level of a trainee on the third day given day 1 and 2 examinations, separately. Cardiac ultrasound examinations were captured with a low-cost probe that captured the 9 degrees of freedom IMU data. Various data augmentation and preprocessing techniques were employed such as the addition of noise, scaling the data, and downsampling. By systematically assessing the performance of each combination of hyperparameters, we have determined the best-performing model that outperforms the mean skill, zero rule classifier. Furthermore, our analysis sheds light on the impact of different data augmentation and preprocessing techniques on model performance. This study proposes a model that has the potential to forecast the skill of a trainee given previous examinations and highlights avenues for future research to enhance performance.