You don't need to prove statistical properties in deep learning models, but if that was something important, that's something you'd be able to do for any classi

You don't need to prove statistical properties in deep learning models, but if that was something important, that's something you'd be able to do for any classifier that has positive predictive ability. If a classifier is good, then there is an underlying association in the data. With astrology, there's no association in the data, unless you do something weird like constrain the size of your sample, systematically exclude people, or are measuring a different effect. It's really totally absurd to use the methods of science on astrology. Just think about it, you have some data, and there's an effect you'd like to figure out. For whatever is going on in your sample, your explanation is "because the position of X on the day you were born?". It's a huge violation of occam's razor, and skips every plausible explanation for the most impossible. No one that seriously does "data science" could approach a question, and come up with astrological answers, unless it was purely an exercise in B.S., with the express goal of creating content for people interested in astrology to consume. "Data scientists" spend years honing and refining their skills asking and answering questions, it's just incongruent with that set of skills and methodology.