An unsupervised machine learning algorithm is developed to evaluate the rail track stability and detect potentially unstable track segments using on-board Doppler LiDAR velocity measurements. The measurement system includes four Doppler LiDAR velocity sensors that can record small variations in the lateral and vertical velocities of the left and right rails. Extensive tests are conducted in a controlled environment on the high-tonnage loop (HTL) at Transportation Technology Center (TTC). The system is mounted on a loaded car measuring left and right rails’ velocities at various forward speeds. To evaluate the performance of the model, a weak lateral condition is simulated in a section of the High Tonnage Loop (HTL) at the TTC. The results show a promising detection of the small amount of gage widening that occurs in the section of the track that is doctored to have weak lateral resistance.