Vista Academy · Machine Learning
Learn to evaluate rare-event classifiers when missed fraud, false alarms and operational costs matter more than a headline accuracy score.
Complete both the activity and MCQ to unlock the next concept. You can revisit any unlocked concept.
Concept 1 of 8
Rare events can be the most important class
A classification dataset is imbalanced when one label appears far more often than another. In fraud detection, legitimate transactions may make up 99% of records while fraud makes up only 1%. The minority class can still carry the highest business risk.
Always inspect class counts and proportions before choosing a model or metric. A model that ignores the rare class may look successful while missing the problem you were hired to solve.
Transaction class distribution
The minority class is only 1% of this hypothetical dataset. The bars use a relative scale so the small class remains visible.
Use the controls to inspect the concept. Exploring the lab does not itself count as passing the activity or MCQ.
A payment dataset has 9,900 legitimate transactions and 100 fraudulent transactions. What is the fraud prevalence?
Which dataset is most clearly imbalanced?
Educational Demonstration. All interactive metrics and fraud scenarios are illustrative; they are not predictions from a trained or validated model.