This project deals with the forecasting of bitcoin prices, using various features such as GPU pricing, US Inflation Rate, and gold prices. The objective is to create a regression model that can use these features to determine the estimated price. Data is retrieved from various sources and then joined together to create a tabular data format.
I think the idea of trying to replace missing data is really interesting, by using the median of the surrounding data points to try to create an estimate of the data. I also think the new features derived from the original five variables is also interesting. In particular, I like the auto-regressive feature, because it attempts to capture the time-series format of the data. I also like the plots used to compare the predicted and true prices, to help visualize the errors of the model.
One thing that I would explore is how effective different models would perform for regression. For example, random-forest regression could be used on the tabular data rather than a simple linear regression. Furthermore, the distribution of some of the features used seems to hover around 0, this makes me question whether or not these features will be useful. I think that it may be better to have features that are uniformly distributed, such that you aren't biased towards the most frequent features. Furthermore, I would add a time series plot, which would help readers visualize the results of the data better. Finally, I think that the regression problem maybe a little too difficult, but a more feasible problem can be whether or not the price of bitcoin will increase/decrease the following day. This would be a classification problem, which is simpler than a regression problem.
Overall, I think the group has done great work so far, and I think the visuals are very helpful for readers, and I am excited to see what comes next.
This project deals with the forecasting of bitcoin prices, using various features such as GPU pricing, US Inflation Rate, and gold prices. The objective is to create a regression model that can use these features to determine the estimated price. Data is retrieved from various sources and then joined together to create a tabular data format.
I think the idea of trying to replace missing data is really interesting, by using the median of the surrounding data points to try to create an estimate of the data. I also think the new features derived from the original five variables is also interesting. In particular, I like the auto-regressive feature, because it attempts to capture the time-series format of the data. I also like the plots used to compare the predicted and true prices, to help visualize the errors of the model.
One thing that I would explore is how effective different models would perform for regression. For example, random-forest regression could be used on the tabular data rather than a simple linear regression. Furthermore, the distribution of some of the features used seems to hover around 0, this makes me question whether or not these features will be useful. I think that it may be better to have features that are uniformly distributed, such that you aren't biased towards the most frequent features. Furthermore, I would add a time series plot, which would help readers visualize the results of the data better. Finally, I think that the regression problem maybe a little too difficult, but a more feasible problem can be whether or not the price of bitcoin will increase/decrease the following day. This would be a classification problem, which is simpler than a regression problem.
Overall, I think the group has done great work so far, and I think the visuals are very helpful for readers, and I am excited to see what comes next.