Hello there. Just a detail about this challenge. I didn't quite understand why you're using variables so that the R squared is so low in the linear regression of the solution. I'm not really used to predictive, but it appears to me that the best model to explain the correlation should have a higher R squared (I found one which is about 0.67 (adjusted) whereas the "solution" one is around 0.32). And, as i read the topic i didn't found anyone underlying this fact, so i'm a bit confused.
Else, good challenge, very interesting
edit : watched my basics, got my misunderstatement