# Classification for choice data

It is essentially a choice modelling problem, but hopefully can be addressed by classification.

Suppose one needs to choose a route to drive to work among many candidates in his mind. These candidates have been summarized with certain features e.g. distance, cost, and number of red lights en route.

We have many individuals like him. How do we build a classification model to make predictions based on the route feature?

My thought is to label every chosen route as 1 and not chosen routes as 0. Therefore for each individual, there is a 1 and many 0s. But to each individual, their candidates are not at the same length/having the same impedances.

How do I take this into account? By a scaler such as MinMaxScaler to standardise everyone? or if there is a way to take the individuality of choice sets into account in a classification model?

• Do you have to optimize choices for many connected individuals? Is it a real-time study or not? (means if you can change the route at any moment) Commented Oct 20, 2022 at 9:33
• No, it is not a real-time study. Just many people have many routes for different origins and destinations.
– GDI
Commented Oct 20, 2022 at 20:02
• It is not real-time, and the routes have no traffic, right? It is important to know whether or not individuals could alter results if they are disturbing each other part of their common routes. Commented Oct 21, 2022 at 20:08
• Consider no congestion. I am aiming to have a comprehensive way to take into account the individuality of the individual choice set. The traditional choice model allows individuals to consider within their alternatives and use log-likelihood to calculate the best-fitted parameters. The classification however considers all options for all indiviuals together. Is there a way to reproduce the individuality of the sets in some way? That is what I m thinking.
– GDI
Commented Oct 22, 2022 at 8:24

## 1 Answer

The problem seems too complex for being solved in one shot.

I suggest first starting with a small case of 3 routes and trying to optimize them correctly. Then you can increase the routes until reaching the totality.

Then, you can try algorithms like a random forest to explore many scenarios and find good solutions.

https://stackabuse.com/random-forest-algorithm-with-python-and-scikit-learn/

But I'm not sure if an optimization solution like Genetic Algorithms is better.

https://github.com/Yaaximus/genetic-algorithm-path-planning

If there is some sample data, it would be easier to know which solution could work best.

• Does it answer your question? If not, please let me know. Commented Oct 29, 2022 at 14:53