Final peer review #129
Description
Goal: Predict rating score on each course based on students’ evaluation on each specific aspect of the course.
Merits:
- Clear organization. The bold fonts really help the readers to capture important points and keep engaged.
- Interesting to find out that Eng students grade instructor and course based on completely different features.
- Excellent utilization of class materials
Concern:
- You mentioned that your dataset contains 267 different courses. Is it possible that for different subjects (or some other metrics such as course level like the one you’ve implemented), students value different aspects of the course. I noticed that you didn't interpret the results on the course level feature you've included, which would be cool if you did.
- Highly correlated features pose the danger of Multicollinearity, which means some of your columns can be replaced by a linear combination of other columns. As you know, one of the most important assumption for an unbiased and consistent results in linear regression is the independence of regressors. It seems like that you realized this point in section 2.1.1 and explored more in section 2.1.4 by using only top k important features. But I personally think it would be better to decide what columns to leave by looking at what columns can be almost predicted by other columns.
- For the fairness part, I think you should be paying attention to ethnic issue regarding the instructors that are being evaluated instead of the students. However, I do understand it might be hard to have information regarding that.
- Maybe consider giving a more concrete conclusion on the point "used as reference for instructors", such as "XXX should be prioritized for course at level X".