Abstract:Transport policy assessment often involves causal questions, yet the causal inference capabilities of traditional travel behavioural models are at best limited. We present the deep CAusal infeRence mOdel for traveL behavIour aNAlysis (CAROLINA), a framework that explicitly models causality in travel behaviour, enhances predictive accuracy, and maintains interpretability by leveraging causal inference, deep learning, and traditional discrete choice modelling. Within this framework, we introduce a Generative Counterfactual model for forecasting human behaviour by adapting the Normalizing Flow method. Through the case studies of virtual reality-based pedestrian crossing behaviour, revealed preference travel behaviour from London, and synthetic data, we demonstrate the effectiveness of our proposed models in uncovering causal relationships, prediction accuracy, and assessing policy interventions. Our results show that intervention mechanisms that can reduce pedestrian stress levels lead to a 38.5% increase in individuals experiencing shorter waiting times. Reducing the travel distances in London results in a 47% increase in sustainable travel modes.

What problem does this paper attempt to address?

The paper primarily aims to address the following two key issues: 1. **Lack of Causal Analysis in Traffic Behavior Research**: Traditional traffic behavior models can predict human travel choices but have limitations in explaining causal relationships. This means that these models struggle to distinguish which factors truly influence people's travel decisions and which are merely correlated without having a direct causal effect. 2. **Challenges in Evaluating Policy Intervention Effects**: Decision-makers need to understand the potential impacts on the traffic system after implementing specific policies. This requires models that can not only make predictions but also perform counterfactual analysis (i.e., how the system would perform under hypothetical interventions). To address the above issues, the authors propose a new framework called "CAROLINA" (Causal Analysis and Reasoning for Optimal Learning in Navigational Activities). This framework combines the advantages of causal inference techniques, deep learning, and traditional discrete choice models to achieve the following goals: - Explicitly model causal relationships and improve prediction accuracy while maintaining model interpretability. - Propose a method to generate counterfactual models to predict changes in human behavior by integrating normalized flow methods. - Demonstrate the effectiveness of the model in revealing causal relationships, improving prediction accuracy, and evaluating policy interventions. To validate the effectiveness of the proposed model structure, the authors conducted empirical studies using three different datasets, including virtual reality-based pedestrian crossing behavior data, actual travel behavior data of London residents, and synthetic data. These case studies showcase the application capabilities of the CAROLINA framework in different scenarios, such as significantly reducing waiting times by lowering pedestrian stress levels or promoting the use of sustainable travel modes by reducing travel distances in London.

A deep causal inference model for fully-interpretable travel behaviour analysis

Multidimensional Travel Decision-Making: Descriptive Behavioural Theory And Agent-Based Models

Causation versus Prediction: Comparing Causal Discovery and Inference with Artificial Neural Networks in Travel Mode Choice Modeling

Determining causality in travel mode choice

Causal Inference in Travel Demand Modeling (and the lack thereof)

Causal Analysis on Travelers’ Perceptions in Traffic States

Debiased machine learning for estimating the causal effect of urban traffic on pedestrian crossing behaviour

Vision Paper: Causal Inference for Interpretable and Robust Machine Learning in Mobility Analysis

Ex-post Evaluation of Transport Interventions with Causal Mediation Analysis

Causal inference for transport research

Causal Inference and Counterfactual Prediction in Machine Learning for Actionable Healthcare

Causal inference in transportation safety studies: Comparison of potential outcomes and causal diagrams

A causal intervention framework for synthesizing mobility data and evaluating predictive neural networks

Examining Active Travel Behavior Through Explainable Machine Learning: Insights from Beijing, China

Simulation-Based Counterfactual Causal Discovery on Real World Driver Behaviour

A State-Based Inverse Reinforcement Learning Approach to Model Activity-Travel Choices Behavior with Reward Function Recovery

Transport-based Counterfactual Models

Influencing transport-health interactions through incentivised mode switch using new data and models

Unraveling the determinants of traffic incident duration: A causal investigation using the framework of causal forests with debiased machine learning

CausalMob: Causal Human Mobility Prediction with LLMs-derived Human Intentions toward Public Events

Modelling Changes in Travel Behaviour Mechanisms Through a High-Order Hidden Markov Model