In American football, predicting the likelihood of a completed pass is a complex task that requires understanding the dynamic interactions between players on the field. This study leverages player tracking data from the National Football League (NFL) to construct graph-based representations of each play, where players are modeled as nodes and spatial relationships as edges. Traditional Machine learning with the use of Graph embeddings is employed to analyze the interactions in the field at a given frame and predict pass completion probabilities. The results show that graph-based features are worth using to capture nuanced spatial dynamics and player interactions. This approach not only enhances predictive accuracy but also provides interpretable insights into the key factors influencing pass outcomes. The findings suggest potential applications for in-game decision-making, player evaluation, and strategy optimization, marking a significant step forward in sports analytics.