Towards Robust Trajectory Representations: Isolating Environmental Confounders with Causal Learning

TL;DR

This paper introduces a causal learning framework called TrajCL that isolates environmental confounders in trajectory modeling, improving classification accuracy and generalization by removing spurious correlations.

Contribution

It formulates a Structural Causal Model for trajectory learning and applies backdoor adjustment to enhance robustness and interpretability of trajectory representations.

Findings

TrajCL significantly improves trajectory classification accuracy.

The framework demonstrates better generalization across datasets.

TrajCL offers enhanced interpretability of movement patterns.

Abstract

Trajectory modeling refers to characterizing human movement behavior, serving as a pivotal step in understanding mobility patterns. Nevertheless, existing studies typically ignore the confounding effects of geospatial context, leading to the acquisition of spurious correlations and limited generalization capabilities. To bridge this gap, we initially formulate a Structural Causal Model (SCM) to decipher the trajectory representation learning process from a causal perspective. Building upon the SCM, we further present a Trajectory modeling framework (TrajCL) based on Causal Learning, which leverages the backdoor adjustment theory as an intervention tool to eliminate the spurious correlations between geospatial context and trajectories. Extensive experiments on two real-world datasets verify that TrajCL markedly enhances performance in trajectory classification tasks while showcasing superior generalization and interpretability.