Logistic regression

[AniaMatysek]

I've done the part connected with logistic regression. However, some changes may appear in the future as I need to discuss observations and conclusions with Phillipe during our class.

[AniaMatysek]

I've also done the Exploratory Data Analysis. Please let me know if I should add anything, as I wasn't sure what this section should cover.

[PiotMik]

Hey @AniaMatysek - thank you for taking on EDA!

I fixed the bug that was causing the pdf to fail to generate: when writing $ p - value = ...$ the "-" symbol used was not a regular minus, but some symbol not recognized by LaTex. Resolved that, generated the pdf and pushed here.

On the methodology side, I have a couple of points that came to my mind when reading these sections - whether to apply or not, I leave the decision to you: I. EDA

• [ ] I highly recommmend ggplot2 package, to plot those histograms. You would need to pivot_longer the data, to have feature names as columns, but then instead of making h1, ..., h23, you could plot all at once with something like:

  data %>% pivot_longer(...) %>% ggplot(aes(...)) %>% geom_histogram() %>% facet_wrap(~...)

• [ ] What information/takeout do we have from all the plots at the beginning? Do we need all of them if we don't refer to any of them later on? Do we need both violin plots and histograms? Is the plot(AirlinesRaw) even readable with that many variables?
• [ ] This one: plot(AirlinesRaw) kills my laptop. It is computationally heavy, so I put sample_n(...) to have it generated on a sub-sample. And again, do we need it in the way it is presented?
• [ ] I guess what I'm trying to say can be summarised as follows: maybe only plots for some "interesting" observations you made about the data would be sufficient?

II. Logistic Regression

• [x] This is a well written section, I really like what you did here
• [x] I especially appreciate that you included variable interactions. How did you arrive at those specific interactions? Could you describe your thought process (1/2 sentences)? It seems like Class should heavily interact with many other variables as well, right?
• [x] Make sure to leave additional empty line when making bullet points - see the pdf how they rendered.
• [x] I would disagree with your remark about "slight overfit" of your model - it looks good to me. Have a look at the magnitude of standard deviation of test set AUC. Your model has a very stable performance, and in line with the AUC of the training set.