A systematic method for selecting trajectories in data augmentation has been proposed, addressing a critical limitation in mitigating data scarcity for machine learning applications 1. Existing trajectory data augmentation techniques, while demonstrating the efficacy of geometric perturbation, have largely relied on rudimentary random selection. This indiscriminate approach frequently compromises the essential spatio-temporal coherence within datasets, thereby hindering the overall utility and reliability of augmented data. The new research introduces a structured framework designed to intelligently identify and prioritize specific trajectories for augmentation. This refined selection process aims to ensure that synthetic datasets retain their intrinsic spatio-temporal properties, leading to the development of more robust and accurate machine learning models. By enhancing data quality through targeted augmentation, this work offers a path to improve the performance and resilience of AI systems challenged by data scarcity. Such advancements in data preparation directly bolster the security and operational integrity of AI-driven platforms, making them less vulnerable to data-related weaknesses.