"predict whether the final action of a new customer is purchasing (i.e., Revenue is TRUE) or not (i.e., Revenue is FALSE)." - perfect phrasing!
Results section is clearer about relating the performance of the model to the question ("Precision & Recall section" is good!).
Writing!
To improve:
Regarding choosing an optimal threshold, would still like an example of under what circumstance that a business might want higher precision than recall, and under what circumstance a business might want a higher recall than precision.
A little more interpretation of the correlation matrix would be appreciated (eg, key correlations observed?)
Would like to see more ideas for future directions.
Limitations could also stand to be expanded on a bit, but it's covered two major points so that's OK.
It looks like you've only addressed one of three of the points in your peer review feedback (#60). If it's not feasible to address the other two, please indicate why (in the issue is fine), and if it's something you might address later leave the issue open.
Good:
To improve: