Overall I believe this is a pretty well understood dataset in the ML community, so it seems the dataset choice could actually be risky, since it is hard to impress with. Kaggle toy datasets in general are pretty dangerous territory after all.
I like in general the thought put into bias-variance trade-off, the limitations of using linear models, and how it could have affected the results of your analysis/visualizations. There are however, several comments that I would like to make:
-
In general, this is a pretty poorly formatted report. Yes I understand that this is nit-picky, and sorry about that, but a lot of the times the report read like an answer to a problem set, rather than a self-sufficient report. This works against the content of your report since justification of certain techniques, and their relation to the whole analysis could be much easier if using proper formatting, fewer uses of bullet points, and much more indepth explanations of the models used.
-
Overall the complexity of analysis and model in your preliminary analysis is somewhat lacking and simplistic.
Justification for models or need for other models is very high-level, and for a machine-learning related report, one may want to go into more theoretical depth on why, for example, a decision tree can be more flexible without referring to a "sweet spot" which is ambiguous and somewhat informal. This is similar to visuals. I feel that visualization is somewhat unappealing - although appeal isn't the highest priority - and is repetitive. Scatterplots with a line added shouldn't be the best visualization you have. Nor should a simple histogram. I like your analyses, I would advise making them on more meaningful visuals.
It seems you guys have the right idea. Try introducing some more theoretical rigor to the methods you are using, and you guys should be good to go.
Overall I believe this is a pretty well understood dataset in the ML community, so it seems the dataset choice could actually be risky, since it is hard to impress with. Kaggle toy datasets in general are pretty dangerous territory after all.
I like in general the thought put into bias-variance trade-off, the limitations of using linear models, and how it could have affected the results of your analysis/visualizations. There are however, several comments that I would like to make:
In general, this is a pretty poorly formatted report. Yes I understand that this is nit-picky, and sorry about that, but a lot of the times the report read like an answer to a problem set, rather than a self-sufficient report. This works against the content of your report since justification of certain techniques, and their relation to the whole analysis could be much easier if using proper formatting, fewer uses of bullet points, and much more indepth explanations of the models used.
Overall the complexity of analysis and model in your preliminary analysis is somewhat lacking and simplistic.
Justification for models or need for other models is very high-level, and for a machine-learning related report, one may want to go into more theoretical depth on why, for example, a decision tree can be more flexible without referring to a "sweet spot" which is ambiguous and somewhat informal. This is similar to visuals. I feel that visualization is somewhat unappealing - although appeal isn't the highest priority - and is repetitive. Scatterplots with a line added shouldn't be the best visualization you have. Nor should a simple histogram. I like your analyses, I would advise making them on more meaningful visuals.
It seems you guys have the right idea. Try introducing some more theoretical rigor to the methods you are using, and you guys should be good to go.