This project is focused on predicting Bitcoin prices.
There is good justification and detail for the additional datasets as well as the data collection. The datasets for gold pricing and inflation rates seem like relevant features so it’s good that they are included into the model.
It is good that you recognize that your model may be overfitting and suggest some good approaches (ridge regression, classification) to help overcome this issue.
I really do think that this report looks very well done and follows a very logical structure/flow. I’m super impressed by the amount of detail and polish.
I would like a more detailed description of the explanatory variables, especially the feature transformations and what was done when certain terms had missing values. For example, if you have an autoregressive term, what happens when you are at the beginning of the time period where there is not data 30 days ahead.
I think the correlation graphs could be very useful, but you do not seem to have added any interaction terms to counteract this effect.
Although there is some justification for overfitting, I would not immediately jump to overfitting but rather that the type of model (linear model) being used does not fit the structure of the data.
This project is focused on predicting Bitcoin prices.
There is good justification and detail for the additional datasets as well as the data collection. The datasets for gold pricing and inflation rates seem like relevant features so it’s good that they are included into the model. It is good that you recognize that your model may be overfitting and suggest some good approaches (ridge regression, classification) to help overcome this issue. I really do think that this report looks very well done and follows a very logical structure/flow. I’m super impressed by the amount of detail and polish.
I would like a more detailed description of the explanatory variables, especially the feature transformations and what was done when certain terms had missing values. For example, if you have an autoregressive term, what happens when you are at the beginning of the time period where there is not data 30 days ahead. I think the correlation graphs could be very useful, but you do not seem to have added any interaction terms to counteract this effect. Although there is some justification for overfitting, I would not immediately jump to overfitting but rather that the type of model (linear model) being used does not fit the structure of the data.