CausalTAD: Causal Implicit Generative Model for Debiased Online Trajectory Anomaly Detection

2024-05-13

IEEE International Conference on Data Engineering

AI Summary

Trajectory anomaly detection suffers from causal confounding induced by road network bias, where conventional conditional probability (P(T|C)) fails to disentangle spurious correlations between trajectories (T) and contextual features (C). Method: We propose the first do-intervention-based debiasing framework, replacing (P(T|C)) with the causal probability (P(T| ext{do}(C))) as the anomaly scoring criterion. Leveraging do-calculus—introduced here for the first time in this domain—we explicitly model the causal effect of semantic destinations (SD) on trajectories while decoupling and eliminating the road network structure as a confounder. Our approach integrates implicit generative modeling, counterfactual trajectory inference, and flow-based density estimation. Results: Experiments show consistent improvements: +2.1–5.7% AUC on in-distribution data and substantial gains of +10.6–32.7% on out-of-distribution (unseen routes) scenarios, demonstrating significantly enhanced generalization and robustness.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Trajectory anomaly detection, aiming to estimate the anomaly risk of trajectories given the Source-Destination (SD) pairs, has become a critical problem for many real-world applications. Existing solutions directly train a generative model for observed trajectories and calculate the conditional generative probability $P(T vert C)$ as the anomaly risk, where $T$ and $C$ represent the trajectory and SD pair respectively. However, we argue that the observed trajectories are confounded by road network preference which is a common cause of both SD distribution and trajectories. Existing methods ignore this issue limiting their generalization ability on out-of-distribution trajectories. In this paper, we define the debiased trajectory anomaly detection problem and propose a causal implicit generative model, namely CausalTAD, to solve it. CausalTAD adopts do-calculus to eliminate the confounding bias of road network preference and estimates $P(Tvert do(C))$ as the anomaly criterion. Extensive experiments show that CausalTadcan not only achieve superior performance on trained trajectories but also generally improve the performance of out-of-distribution data, with improvements of 2.1% ~ 5.7% and 10.6% ~ 32.7% respectively.

Problem

• Trajectory Anomaly Detection
• Route Preference
• Accuracy Improvement

Innovation

• CausalTAD
• Route Preference
• Anomaly Detection