To predict which auto liability claims will likely go to litigation, an insurer uses data mining to identify known indicators. Which technique would be most appropriate?

To predict which auto liability claims will likely go to litigation, an insurer uses data mining to identify known indicators. Which technique would be most appropriate?

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

Answer: (C)

Predicting which auto liability claims will go to litigation is a classification task because you’re assigning each claim to a discrete outcome (litigation vs. not litigation) based on input indicators such as claim details, prior history, and policy features. In classification, the model is trained on labeled historical data to learn patterns that distinguish claims that end up in litigation from those that don’t, then applied to new claims to estimate the likelihood of litigation. Association rule learning looks for co-occurring items and doesn’t directly output a category for each case. Cluster analysis groups similar claims without using a predefined outcome, so it’s unsupervised and not suited for predicting a label. Regression predicts a continuous value (like cost or days to resolution) rather than a binary decision. Thus, classification best fits the goal of predicting a yes/no litigation outcome.

Predicting which auto liability claims will go to litigation is a classification task because you’re assigning each claim to a discrete outcome (litigation vs. not litigation) based on input indicators such as claim details, prior history, and policy features. In classification, the model is trained on labeled historical data to learn patterns that distinguish claims that end up in litigation from those that don’t, then applied to new claims to estimate the likelihood of litigation. Association rule learning looks for co-occurring items and doesn’t directly output a category for each case. Cluster analysis groups similar claims without using a predefined outcome, so it’s unsupervised and not suited for predicting a label. Regression predicts a continuous value (like cost or days to resolution) rather than a binary decision. Thus, classification best fits the goal of predicting a yes/no litigation outcome.