Week 9: Model Adjustments
Ishan B -
Last week, I faced some unexpected results with the linear regression model. The score of the model was exceptionally low. My main ideas were that either there were too many features so the model was having difficulties fitting to the data or that the data is too far from a linear trend to have any success.
This week, I implemented some other metrics to gain a better understanding of whether the model was truly performing poorly or if it was performing well on other metrics. The most commonly used metric for regression models is root mean squared error (RMSE). RMSE gives a number showing how far the regression is from fitting the data. This is similar to the information that score gives except that you want a low RMSE number. The RMSE from the linear regression was 33303.971964744225. This is an exceptionally large number showing that the linear regression model was truly not fitting the data well. The reasoning for why that is happening is most likely just that the data does not follow a linear trend, which is usually the case for most datasets. Even with that, I believe that training a linear model first is valuable information as it gives a baseline accuracy and number to compare the other model types to.
Next week, I am planning on training a polynomial regression model. I hope that this model provides a lower RMSE and higher score, and I believe that it will as it is likely that the data has some sort of curve to it, meaning that the polynomial model should fit better.