You are evaluating two regression models built to predict residential home values. Model X has a Mean Absolute Error (MAE) of $18,000, while Model Y has an MAE of $25,000. Which model is preferred based on this metric, and what does it tell you about the models' predictions?
Select an answer to reveal the explanation.
Short Explanation
When you're predicting home prices, you want your guesses close to reality. The $18,000-MAE model has smaller average misses than the $25,000-MAE model, so you prefer it. Lower MAE means smaller errors.
Full Explanation
In regression tasks, performance metrics assess how close a model's continuous predictions are to the actual target values. Mean Absolute Error (MAE) is one of the most straightforward and interpretable metrics used for this purpose.
Mathematically, MAE is calculated by taking the absolute difference between the predicted value and the actual value for each data point, summing these absolute differences, and dividing by the total number of data points. Because it uses absolute values, MAE measures the average magnitude of the errors without considering their direction, whether the prediction was too high or too low. It is expressed in the same units as the target variable, in this case dollars, making it highly intuitive for business stakeholders.
Since MAE represents error, a smaller value is always preferred. The lower MAE indicates that, on average, the model's predictions deviate less from the true target values, signifying a more accurate model. Therefore, the model with the $18,000 MAE is preferred over the model with the $25,000 MAE because its average prediction error is smaller.
The claim that the model with the higher MAE is preferred because a higher MAE indicates the model has captured more data variance is incorrect because a higher MAE indicates a larger average error, meaning the model's predictions are less accurate. The claim that the model with the lower MAE should be rejected because a low MAE is a primary symptom of severe model underfitting is incorrect because a low MAE on validation or test data is a sign of good model performance, not underfitting. While a low training MAE combined with a high validation MAE indicates overfitting, MAE itself is not a symptom of underfitting. The claim that neither model can be compared using MAE because MAE is only applicable to binary classification metrics is incorrect because MAE is specifically designed for regression tasks that predict continuous variables and cannot be applied to classification tasks that predict discrete classes.
Exam caveat: lower MAE simply means closer predictions.