Accurate prediction of a vehicle’s short-term trajectory has been an essential focus for effective traffic management, safety, and driving experience, especially in rapidly evolving self-driving technology. This paper proposes a novel approach to predicting a car’s destination within 10 s by considering its current velocity, acceleration, and geolocation data. The proposed model uses LambdaMART with a Light Gradient Boosting Machine (LGBM) to rank potential destinations in decreasing order of certainty while minimizing the average Normalized Discounted Cumulative Gain (NDCG) score of all queries across all rank positions. The usefulness of the proposed model is validated using the Next Generation Simulation (NGSIM) Vehicle Trajectories dataset, which has over 11.8 million trip data points from Southern California. The model demonstrates remarkable accuracy, with the actual destination node consistently placed within the top 2 predictions in 100% of the 30,000 test instances. This approach does have certain constraints, like its dependency on OpenStreetMap, which introduces some error margins when the real destination does not correspond to any nodes in the map. This work is preliminary, and further research is required to enhance the model’s capabilities and address its current limitations.