Ever since the 2024 International Choice Modelling Conference (ICMC) was announced back in 2022, with Chile as the host, I knew I had to go. Who could resist a trip to Chile? Two years later, that decision paid off as I found myself in Puerto Varas, ready to present my second research paper titled: “Generative Latent Class Choice Models: Enriching Smart Card Data with Travel Surveys using Variational Auto-Encoders“ and co-authored with Filipe Rodrigues and Francisco Camara Pereira.
Our study explores the integration of Machine Learning (ML) and Discrete Choice Models (DCM) through the development of the Variational Autoencoder and Latent Class Choice Model (VAE-LCCM). By leveraging the generative capabilities of Variational Autoencoders (VAEs), this model enhances traditional behavioral choice models. It generates synthetic data, imputes missing information, and offers a better representation of heterogeneity. The neural network architecture of VAEs allows for dimensionality reduction and non-linear pattern recognition, leading to improved goodness-of-fit and better out-of-sample generalization for the LCCM, all while preserving its behavioral and economic interpretability.
The conference itself was a great experience. I had the chance to reconnect with old colleagues and friends, meet new peers, and engage with some of the latest advances in the choice modeling field.
You can click here to check the slides!
Leave a Reply