Which data mining technique is typically used when the goal is to predict a discrete category such as 'will go to litigation'?

• A. Clustering
• B. Classification
• C. Regression analysis
• D. Association rule learning

Answer: (B)

Predicting a discrete category, such as whether a claim will go to litigation, uses a classification approach. Classification is a supervised learning task where the target outcome is categorical (for example, litigation vs. no litigation). You train on historical claims that are labeled with the outcome, learn how features map to those categories, and then assign the same category to new cases. This differs from regression, which predicts a numeric value, and from clustering, which partitions data into groups without predefined labels, and from association rule learning, which uncovers relationships between items rather than assigning a class to each instance. In practice, you’d evaluate classification models with metrics like accuracy, precision, recall, or ROC-AUC and use cross-validation to ensure the model generalizes well.