Few Shot Semi-Supervised Learning for Abnormal Stop Detection from Sparse GPS Trajectories

TL;DR

This work tackles abnormal stop detection on long-distance coach routes under sparse GPS sampling and minimal labeling. It introduces Sparsity-Aware Segmentation (SAS) to adaptively partition trajectories, three domain indicators (Temporal Influence Score, Maximum Speed Deviation, Top-$k$ Aggregated Temporal Score) to capture abnormal stops, and LTIGA to smooth indicators via local temporal graphs. These features feed into a global spatial-temporal graph where label propagation, a GCN, and a self-training loop leverage scarce supervision to detect abnormal stops with high accuracy, as demonstrated on real-world data (AUC 0.854, AP 0.866 with just 10 abnormal labels). The approach achieves strong performance and spatial localization while remaining interpretable through the domain-specific indicators, offering a practical, label-efficient solution for ASD in transportation systems.

Abstract

Abnormal stop detection (ASD) in intercity coach transportation is critical for ensuring passenger safety, operational reliability, and regulatory compliance. However, two key challenges hinder ASD effectiveness: sparse GPS trajectories, which obscure short or unauthorized stops, and limited labeled data, which restricts supervised learning. Existing methods often assume dense sampling or regular movement patterns, limiting their applicability. To address data sparsity, we propose a Sparsity-Aware Segmentation (SAS) method that adaptively defines segment boundaries based on local spatial-temporal density. Building upon these segments, we introduce three domain-specific indicators to capture abnormal stop behaviors. To further mitigate the impact of sparsity, we develop Locally Temporal-Indicator Guided Adjustment (LTIGA), which smooths these indicators via local similarity graphs. To overcome label scarcity, we construct a spatial-temporal graph where each segment is a node with LTIGA-refined features. We apply label propagation to expand weak supervision across the graph, followed by a GCN to learn relational patterns. A final self-training module incorporates high-confidence pseudo-labels to iteratively improve predictions. Experiments on real-world coach data show an AUC of 0.854 and AP of 0.866 using only 10 labeled instances, outperforming prior methods. The code and dataset are publicly available at \href{https://github.com/pangjunbiao/Abnormal-Stop-Detection-SSL.git}

Figures (5)

• Figure 1: Illustration of Sparsity-Aware Segmentation (SAS). A segment break is introduced when the spatial distance $D_{ij}$ or time gap $\Delta t_{ij}$ between consecutive GPS points exceeds adaptive thresholds $\lambda_d$ or $\lambda_t$, respectively. These thresholds are dynamically computed using Equation \ref{['eq:adaptive_thresholds']}.
• Figure 2: Spatial overlays of predicted vs. ground-truth abnormal stops under three label budgets. Prediction improves with more supervision—from failure (a) to moderate (b) to high-quality detection (c).
• Figure 3: Map and street view of predicted stop at (116.3025, 39.9584).
• Figure 4: Map and street view of predicted stop at (116.3044, 39.8957).
• Figure 5: Comparison of segmentation distributions under SAS and fixed-length segmentation