AI-900 Practice Question: Describe fundamental principles of machine learning on Azure
What is the 'mean absolute error' (MAE) metric used to evaluate in machine learning?
⚠ Common exam trap
Test-takers frequently confuse MAE with classification metrics like accuracy or confidence, or assume it involves thresholds, when in fact MAE is strictly a regression metric measuring average absolute error without any threshold or comparison to random chance.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
The average absolute difference between regression model predictions and actual values
Mean Absolute Error (MAE) is a regression metric that calculates the average of the absolute differences between predicted and actual values. It measures how close predictions are to the true outcomes, with lower values indicating better model accuracy. In Azure Machine Learning, MAE is commonly used to evaluate regression models like linear regression or decision forests.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
The average confidence percentage of classification predictions
Why it's wrong here
Confidence percentage refers to the predicted probability output of a classification model, such as a 0.85 chance of class membership, and is evaluated with calibration or Brier score. MAE only applies to regression tasks where the target is continuous, measuring the average distance between predicted numbers and actual numbers. A disproportionately confident classifier (e.g., always predicting 0.99) can have perfect confidence calibration yet be entirely wrong about which class is present, showing confidence is not error magnitude.
- ✓
The average absolute difference between regression model predictions and actual values
Why this is correct
Mean Absolute Error (MAE) is computed as the sum of the absolute differences between each regression prediction and its actual value, then divided by the sample size. This yields the average error magnitude in the same units as the target, and lower MAE indicates better fit. Unlike squared-error metrics, MAE does not penalize outliers disproportionately, so it is a robust, easily interpretable measure of typical prediction error.
- ✗
The proportion of model predictions that deviate from expected values by more than a threshold
Why it's wrong here
This phrasing describes a threshold-based error count, such as the proportion of points whose absolute error exceeds a specified tolerance (e.g., a 10% error bound), not the average error size. MAE sums the absolute deviations for every prediction and divides by the number of samples, so it reflects the typical magnitude of error across the whole dataset, not the frequency of large failures. A model with zero threshold violations could still have a high MAE if many small errors exist.
- ✗
How much the model's predictions differ from random chance
Why it's wrong here
Measuring how predictions differ from random chance is the domain of ROC/AUC, lift, or information-gain statistics, which compare model ranking to a baseline coin-flip or prevalence rate. MAE instead quantifies the raw difference in units of the target variable between each predicted value and its true value, making it a scale-dependent accuracy measure. A model could perform far better than random chance (high AUC) yet still have a high MAE because AUC ignores calibration and absolute error.
Go deeper
Related to this question
Learn chapter
Machine Learning Core Concepts
Key term
Machine learning
Machine learning is a branch of artificial intelligence where computers learn patterns from data to make decisions or predictions without being explicitly programmed for every task.
Key term
Regression
Regression is a type of machine learning algorithm that predicts a continuous numeric output based on input data, used to model relationships between variables.
About these practice questions
One of 985 original AI-900 practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI-900 practice question is part of Courseiva's free Microsoft certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI-900 exam.