 Finally, the last step is to evaluate variable importance. So if we just do our summary, like we did in classification and run that, it once again prints out this giant set of text. But it's telling us that we ran a regression model, it's telling us what our features were, and then it gets into variable importance. And once again, the second variable importance is what we are going to be most interested in. So in classification, we say a mean decrease in accuracy, but in variable importance, we look at the mean increase in root mean squared error. So this means as we shuffle the data set, one variable at a time, we see how much did our error increase after we shoveled that data because we've broken that relationship. The higher the increase in error, the more important that relationship is to our model. And so we can see that the total square footage is the most important variable when we shuffled it. It increased the root mean squared error by 13.9 cubic feet of natural gas, which we can see is quite significant. And then it goes down. It is interesting to note that we've got some negative values here. And so these variables when shuffled, actually on average led to decreases in root mean squared error. So mean increase, negative means a decrease, which means the model got better when we shuffled it, which means these three variables are completely unimportant to the model to the point where randomizing them led to at times no change, but oftentimes led to improvements in the model. And so this is how you can use this variable importance to make decisions on whether or not to keep certain models. So if we remove these three variables, stove oven, number of fridges, LG, TIN, four, then perhaps we actually might improve our accuracy because we're not giving it these extraneous variables that don't have anything to do with the analysis. If we continue down, we once again see all of these variable importance is less than in classification because there's less ways to measure it, but still quite a few. It gives us the out of bag root mean squared error, which you'll notice is different than the root mean squared error that we got above, which was 374, missed it. This is 387. So usually you will see a difference here. This out of bag evaluation is an internal random forest measure that it does automatically. And so it's still done on the training set, but is often sort of can occasionally be used and measure the error. If you don't have a large enough test set to really get that accurate, to do that prediction, you can do out of bag error, which is a very similar tool, but sort of works internally to random forest. And then there's things about the actual tree composition and different attributes of our variables. And finally, here's our root mean squared error out of bag evaluation as the trees, number of trees increase. But similar to classification, when looking at this, this will be the critical piece for our analyses within this class.