The idea is again very much marketable. Lot's of real estate companies/customers would be interested.
Exploratory data analysis is done and respective graphs are plotted. Which gives us an idea of whih columns would be most relevant and which ones dont actually determine the house prices.
The challenge of using AR model was discussed. It is a very shrewd observation that if something out of order happened during past then we cannot really use old years data for regression unless you use other financial factors to find out next recession. Which would be a complicated project in its own.
Feature discussion done at sufficient length and discussed which new features would be useful.
Ridge , lasso and L2 regressions were used to get an initial estimate of success rate.
Improvements:
Data corruption cleaning , although mentioned but still data seems to be unclean, as you mentioned there are a lot of 0 priced entities.
The success rate in mse seems abnormal, probably because of unclean data ?
The report is slightly long and some of the information given in plots and text is redundant, i am sure it could be reduced down to 4 pages or so.
Underfitting has been discussed but i was left wondering for a moment why overfitting won't be an issue but underfitting will. An explanation of that would have been really good.
Overall, i think it is a brilliant idea and team has some really good analysis done and is moving towards achieving a final result. Thanks !
plus points
Improvements:
Overall, i think it is a brilliant idea and team has some really good analysis done and is moving towards achieving a final result. Thanks !