This report aims to predict the score that a film will receive on IMBd, a well-known movie rating database, using information about the cast and production of the movie. To perform their analysis, the group utilizes a dataset found on Kaggle that contains information about ~4800 movies, scraped from the IMBd website.
Three Things I Enjoyed
-
I really appreciated all of the approaches that you made use of--in particular, I found it to be very insightful that you not only identified your analysis as a regression problem, but also as a classification problem. Determining whether a movie is "good" or "bad" seems just as important, if not more important, than scoring a movie as "8.8" rather than "8.9", as you guys note.
-
I think the inclusion of your correlation matrix among the features was nice--it was a good way to visualize the final relationships among your variables, especially after the many feature transformations you guys took on.
-
You guys did a really stellar job of framing your problem--bringing into context the important of casting and producing in this competitive entertainment industry (with Netflix, HBO, etc). Furthermore, the dataset that you guys worked with was really appropriate for your problem, and I felt like none of the transformations and models were "forced"--as in your dataset lent itself naturally to the various approaches you took.
Three Areas of Potential Improvement
-
In your dimensionality reduction process, you explained how you made use of an ordinary least squares model to select the features that were likely to be most significant. I wonder why you did not make use of a LASSO model for your feature selection instead, as we have learned that the LASSO problem encourages sparsity and allows for more interpretability. While you did use LASSO for your actual model, it may have been good to use as a "first-step" feature-selection model, as was stated was common by Professor Udell in lecture.
-
This is more of a presentation issue, but I found the headings to be kind of unclear. The sub-sub-headings seemed to be only slightly smaller than the other sub-headings, so I had difficulty identifying where one sub-sub-section ended and another began.
-
Given the large number of models you created, supervised (regression and classification) and unsupervised (k-means, PCA, etc.), I feel like the analysis of your results was somewhat lacking. For each method, you explained the model and reasoning well (though I'm not sure it was completely necessary to go so in-depth?) but the results summary seemed to just be a couple of sentences about the error metrics. A more comprehensive analysis about the performance of the unsupervised vs supervised methods you used, as well as how the end users (filmmakers and such) could actually make use of the model despite other restrictions (e.g. we know that having a stellar cast is important, but what if there is no budget for it?) would have been appreciated. The team did hint at this at the end, but fleshing it out more would be good.
Overall, solid work, and I hope you enjoyed learning about what makes a quality movie!
This report aims to predict the score that a film will receive on IMBd, a well-known movie rating database, using information about the cast and production of the movie. To perform their analysis, the group utilizes a dataset found on Kaggle that contains information about ~4800 movies, scraped from the IMBd website.
Three Things I Enjoyed
I really appreciated all of the approaches that you made use of--in particular, I found it to be very insightful that you not only identified your analysis as a regression problem, but also as a classification problem. Determining whether a movie is "good" or "bad" seems just as important, if not more important, than scoring a movie as "8.8" rather than "8.9", as you guys note.
I think the inclusion of your correlation matrix among the features was nice--it was a good way to visualize the final relationships among your variables, especially after the many feature transformations you guys took on.
You guys did a really stellar job of framing your problem--bringing into context the important of casting and producing in this competitive entertainment industry (with Netflix, HBO, etc). Furthermore, the dataset that you guys worked with was really appropriate for your problem, and I felt like none of the transformations and models were "forced"--as in your dataset lent itself naturally to the various approaches you took.
Three Areas of Potential Improvement
In your dimensionality reduction process, you explained how you made use of an ordinary least squares model to select the features that were likely to be most significant. I wonder why you did not make use of a LASSO model for your feature selection instead, as we have learned that the LASSO problem encourages sparsity and allows for more interpretability. While you did use LASSO for your actual model, it may have been good to use as a "first-step" feature-selection model, as was stated was common by Professor Udell in lecture.
This is more of a presentation issue, but I found the headings to be kind of unclear. The sub-sub-headings seemed to be only slightly smaller than the other sub-headings, so I had difficulty identifying where one sub-sub-section ended and another began.
Given the large number of models you created, supervised (regression and classification) and unsupervised (k-means, PCA, etc.), I feel like the analysis of your results was somewhat lacking. For each method, you explained the model and reasoning well (though I'm not sure it was completely necessary to go so in-depth?) but the results summary seemed to just be a couple of sentences about the error metrics. A more comprehensive analysis about the performance of the unsupervised vs supervised methods you used, as well as how the end users (filmmakers and such) could actually make use of the model despite other restrictions (e.g. we know that having a stellar cast is important, but what if there is no budget for it?) would have been appreciated. The team did hint at this at the end, but fleshing it out more would be good.
Overall, solid work, and I hope you enjoyed learning about what makes a quality movie!