Counterfactual Prediction in Transport Modelling - Ricardo Silva & Charisma Choudhury

Explore counterfactual prediction in transport modeling through this 24-minute workshop from the Alan Turing Institute. Delve into the limitations of traditional prediction algorithms and discover how causal inference can enhance decision-making capabilities. Learn about the outcomes of a Turing Institute challenge focused on addressing methodological challenges in counterfactual prediction. Gain insights into the application of these methods for decision support during the COVID-19 pandemic. Examine topics such as causality in transport modeling, disrupted exits, potential uses of counterfactual prediction, and flow-based featurization. Analyze examples of parameterization, output comparison, and out-of-sample evaluation techniques. Understand the importance of moving beyond expectations and comparing scores in predictive modeling. Conclude with a discussion on exit-count distributions and the broader implications of this research for the field of transport modeling and decision support systems.

Intro
Causality in Transport Modelling
This Talk Problems on interest
Disrupted Exits Will Not Fit Natural Regime
Potential Uses
Causal Prediction in Our Setup
Examples of Related Work
Counterfactual Mapping in the Underground A Flow-Based Featurisation . Covariates here are the state of the system at the moment of the disruption, so all random variables are conditioned on the past of the system.
Example of Parameterisation
Example of Output
Moving Beyond Expectations
Comparing Scores
Out-of-Sample Evaluation
Comparison
Log-likelihood of Disrupted Exit Counts Log-scale
Exit-Count Distributions
Conclusions