Exploring common regression metrics

- [Instructor] Metrics are key indicators of whether or not your model is well performing, or whether you'll need to tweak the hyperparameters to continue your training iterations. Today, we'll cover several metrics reserved for regression problems. R squared, mean squared error, root mean square error, and mean absolute error. Let's talk about them now. We've used R squared throughout the course to evaluate our home cost predicting model. If you recall, R squared calculates the difference between the actual values and the predictions made by the model. The distances between the actual and predicted values are called residuals. Residuals are key to determining the performance of a regression model. R squared values land between zero to one. Values closer to one indicate a model that has a better fit between predicted and actual values. R squared is more of a relative measure, while mean square error and root mean square error are absolute measures. Means square error, or MSE, is an absolute number of how much your predicted results deviate from the actual number. Root means square error, or RMSE, is the square root of MSE, making MSE easier to interpret. Mean absolute error, or MAE, is much like mean square error, however, mean absolute error is a more direct representation of the sum of error predictions. It takes the sum of the absolute value of errors. Simply put, it's the sum of all the differences between the actual and predicted values, divided by the total number of predictions in the dataset. You'll use mean absolute error when you want to know how close the predictions are to the actual values on average. Let's calculate these metrics for the home predicting model and interpret what these metrics tell us about our model. The R squared score function is used to calculate R squared. You'll pass in the actual values stored in Y test and the predicted values stored in y_xgb_pred_test. When we calculate the score, we see it lands right at 77%, which means the model is correct or accurate 77% of the time. We'd want this number to be higher before deploying this model to production. Now let's look at the mean absolute error. You'll pass in the actual and predicted values. Let's scroll down here. Here's the mean absolute error. We're using this mean absolute error function and notice the number comes to 36,584. How do you interpret the score? Both the mean absolute error and the root mean square error range from zero to infinity. Lower mean absolute error values indicate that the model is correctly predicting, while larger mean absolute error values indicate the model has poor predicting capabilities. Mean square error and root mean squared error are up next. Again, we're passing in the actual, right here, and predicted values. Root means square error helps to determine if there are any large errors in your model. Did your model predict many values that were significantly higher or lower than the actual values? The root mean square error score will tell you that. First we'll need to determine the mean square error, and that's what we do here, using this mean squared error function. First, we'll need to determine that and then we can calculate the root mean square error using this mean squared error function. We have our values here at the bottom. The first is the mean square error and the second is the root mean square error. We can see that the root mean squared error value is larger than the mean absolute value. This means that there are some large errors in the dataset, which is in line with the need to improve the accuracy which landed at 77%. We'll need to improve that accuracy before deploying the model. Now that you understand how to evaluate regression metrics using R squared, mean absolute error, mean squared error, and root mean square error, let's move on to feature importance.