Outstanding post Zoe, I am glad that this nice project has been published!
I think it motivates and demonstrates essential machine learning and classification concepts very neatly.
However I can't resist to warn the general public, that – as Zoe pointed out in the first sentences – poisonous mushrooms are often deadly, so be extremely careful even with a high accuracy model.
To take into account of this asymmetry of classes, a natural enhancement might be to tweak the UtilityFunction of the model:
https://reference.wolfram.com/language/ref/UtilityFunction.html
Similar to the examples in Etienne Bernard's book:
https://www.wolfram.com/language/introduction-machine-learning/classification/
Or one can be just more conservative in their mushroom picking practices :)