Last year I worked with a start-up called Scentronix, which makes customized perfumes. The company made a device which can quickly blend fragrances on the spot based on 200 most commonly used ingredients in the business. Apparently, you can make most of the available perfumes with these base ingredients. The recipes are a bunch of known well-received recipes, and it will also blend new random recipes (with certain constraints like don't add too much of an overwhelming base scent). They sample feedback from people and train a collaborative filtering classifier. It is a hard problem, and the data is very noisy, but we did start to get better-than-random results at some point. https://scentronix.com/

You could do something similar to how they trained a ML model to find antibiotics compounds: https://www.cell.com/action/showPdf?pii=S0092-8674%2820%2930.... First, train a deep learning model to learn a representation of molecules from their molecule structures. Then feed in the thousand or so known compounds that produce pleasant or unpleasant smells as training data with some score of "pleasantness". We can then use this model to quickly score millions of compounds and select candidates to test.

I love the explanation.

I guess a better way would be to monitor their brain activity. That being said I consider most of the perfumes polution.