dancyfang
commented
7 years ago Thanks for your review! We will be more careful about our writings next time. Maybe a pairwise correlation plot will be better than this scatter plot. We deleted FaciFAR because its strong correlation to the other two variables will make it redundant and may lead to overfitting. We didn't choose PCA because after projection, the meaning of each predictor will be unexplainable.
Strengths:
What the team has achieved so far is impressive. I like the fact that you already developed several models and compared results. Such preliminary results provided good guidance for future steps.
I love the idea of finding more information about certain areas in the city. I think it would
Some concerns
you should be more careful about writing and proofread the report before you submitted it. There are some obvious errors and it's hard to read (e.g. at the bottom of page 2, X and Y both have AssessTot listed).
when you mentioned that you deleted one feature because it correlates with other two variables. Is there a specific reason why you kept the others instead of the deleted one? Since correlation has no direction, it would make sense to keep either if your solution to correlated features is to delete one of them. Also, a better way to deal with correlated features is to do PCA and use composite components instead of individual features.
-Why did you exclude the zip code feature in the first model? Also, direct comparisons between the two models should be included.