Mastering Machine Learning on AWS
上QQ阅读APP看书,第一时间看更新

Evaluating regression models 

Unlike the Naive Bayes classification model, the regression model provides a numerical output as a prediction. This output can be used for binary classification by predicting the value for both the events and using the maximum value. However, in examples such as predicting a house value based on regressors, we cannot use evaluation metrics that rely on just predicting whether we got the answer correct or incorrect. When we are predicting a numerical value, the evaluation metrics should also quantify the value of error in prediction. For example, if the house value is 600,000 and model A predicts it as 700,000 and model B predicts it as 1,000,000, metrics such as precision and recall will count both these predictions as false positives. However, for regression models, we need evaluation metrics that can tell us that model A was closer to the actual value than model B. Therefore, in this section, we will present three metrics that are used for such numerical predictions.