Proficiency in foreign languages is crucial for effective global operations, and proficiency can be gained from training courses. In this work, we study the influence of various training courses by analyzing a 9,436-row dataset, including factors such as language, test timing, and their effect on an individual's language test performance. Both logistic regression algorithms and neural networks were employed to analyze the factors influencing language test scores. The key findings reveal that while the neural network model slightly outperformed the logistic regression model in terms of positive class recall (0.69 vs. 0.61), the logistic regression model is preferred due to its greater interpretability and generalizability. Notably, the logistic regression model identified higher initial reading scores and frequent testing as significant predictors of score improvement. Importantly, the logistic regression model achieved an accuracy of 0.78, which significantly surpasses the trivial model's accuracy of 0.51. This study contributes to the field of predictive analytics in language proficiency and provides actionable insights for future research and policy-making in defense language programs. The interpretable logistic regression model can help inform the design of more effective language training courses and testing strategies to enhance global operations.