After presenting a data-driven modeling prototype at the 2016 Innovations in Travel Modeling Conference, some asserted that it could not be used in forecasting. Nothing could be further from the truth. We admit, though, that forecasting would perhaps not be in the way we have traditionally done so. Josie Kressner and I put together a presentation to illustrate the wide range of approaches that can be used with data-driven models (or any other type of model, for that matter) at the TRB Transportation Planning Applications Conference this week. Cramming it all into a 15-minute presentation necessitated firehosing the audience with ideas, but few concrete examples were included. So we are going to write a series of blog posts here that discuss each of them in more detail.
The eight approaches we presented include:
- Embedding traditional behavioral models
- Model coupling
- Distribution shaping
- Evolutionary models
- A/B testing
- Agent-based models
- Machine learning
By the time we get done writing all of these up we might think of others. Comments or ideas are always welcome. Each post in this series will be in the forecasting category.