AI-900 Practice Question: Describe fundamental principles of machine learning on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  ROC-AUC is always better than accuracy regardless of the use case

  Why it's wrong here

  No evaluation metric is universally superior; the choice depends on the problem context. Accuracy is perfectly valid when classes are balanced and the cost of false positives and false negatives is symmetric, because it directly measures the expected proportion of correct decisions. ROC-AUC is threshold-agnostic, so it does not reflect performance at a specific operating point, and in highly imbalanced scenarios it can be overly optimistic compared to precision-recall curves or F-beta. Therefore, claiming ROC-AUC is always better than accuracy ignores the importance of business constraints and the interpretation of the actual deployed threshold.

• ✓

  A threshold-agnostic metric that measures discrimination ability — better than accuracy for imbalanced classes

  Why this is correct

  ROC-AUC evaluates how well a model separates positive and negative classes by measuring the probability that a randomly chosen positive instance is ranked higher than a randomly chosen negative instance, across every possible decision threshold. Because it aggregates the true positive rate and false positive rate over all thresholds, it does not require picking a single cutoff and is therefore threshold-agnostic. In imbalanced datasets, accuracy can be misleadingly high by simply predicting the majority class, whereas ROC-AUC reflects the model's ranking ability and stays informative even when one class dominates.

• ✗

  A metric specifically for measuring multi-class classification across more than two classes

  Why it's wrong here

  ROC-AUC is fundamentally a binary classification metric: it plots the trade-off between true positive rate and false positive rate as a single discrimination threshold is varied. For multi-class problems, it requires a one-vs-rest decomposition or averaging scheme (e.g., macro or micro), and even then it loses the intuitive threshold interpretation. Other metrics like log loss or quadratic weighted kappa are more naturally designed for multi-class evaluation, so this option mischaracterizes ROC-AUC as a specialized multi-class metric.

• ✗

  An evaluation metric only applicable to models trained on Azure Machine Learning

  Why it's wrong here

  ROC-AUC is a purely statistical measure computed from predicted scores and ground-truth labels; it does not depend on where a model was trained. Any platform or framework that supports classification can calculate it, from Scikit-learn to PyTorch to Azure Machine Learning. Azure ML merely provides automated scorecards that include ROC-AUC, just as it does for accuracy, F1, and precision. Thus, limiting it to Azure-trained models confuses the metric itself with the tooling that reports it.

Go deeper

Related to this question