Unveiling implicit path outlier deviations via foundational linear inverse optimization in mobility networks

This study introduces a new framework for inferring hidden link costs and detecting structural outliers in advanced mobility networks, where observed trajectory data include irregularities and non-optimal paths under available link costs. These networks, increasingly characterised by the integration of connected and autonomous vehicles and diverse mobility services, generate data that challenge traffic analysis. Our method is based on the inverse shortest paths problem formulated through linear programming, extending it to infer implicit cost structures from noisy paths. By reframing outlier detection as the identification of structural deviations through inverse inference, rather than direct anomaly identification, our approach reveals structural inconsistencies critical for maintaining reliable mobility systems. Evaluated on synthetic and real networks, our method consistently achieved significantly better recall rates compared to benchmark methods on large-scale datasets, particularly in detecting spatially close outlier deviations. Our proposed approach supports real-time applications and provides a robust foundation for AI-based outlier detection.