This study explores the development and application of gray box models in predictive learning analytics using clickstream data. Although learning analytics may help enhance educational processes by using predictive models, the balance between the accuracy of black box models and the interpretability of white box models presents a significant challenge. This research introduces gray box models, which aim to integrate the interpretability of white box models with the accuracy of black box models. Two primary approaches are investigated: incorporating educational theories into machine learning algorithms, and applying Explainable Artificial Intelligence (XAI) techniques to enhance model understanding. The study proposes to explore the use of advanced machine learning techniques, such as Random Forests, Gradient Boosting, and Neural Networks, alongside XAI methods like SHAP and LIME, to predict student performance and provide actionable insights. Gray box models may offer a balanced solution, maintaining high predictive accuracy while ensuring interpretability, thus supporting more informed and effective educational decision-making.