GCP Generative AI Leader Practice Question 2: Labeled Data Learning Approach
Google Cloud Generative AI Leader · Test 1 · question 2 of 10
An insurance operations manager has five years of historical claims. Each claim is tagged by an investigator as either "fraudulent" or "legitimate". She wants a model that predicts which new claims are likely to be fraudulent, so investigators can review those first.
Which machine learning approach fits this data and goal?
Show answer and explanation
Answer:
The key fact is that the data is labeled. Every historical claim comes with the correct answer, "fraudulent" or "legitimate", as assigned by an investigator. Supervised learning trains on exactly these pairs of inputs and correct outputs, learns the relationship between them, and then predicts the label for new, unseen claims. Predicting a known category from labeled history is the textbook supervised case.
Option A, unsupervised learning, works on unlabeled data. It finds clusters or anomalies without being told what the right answer is. It helps when no labels exist, but here the labels are the most valuable asset, and ignoring them would throw information away.
Option B, reinforcement learning, trains an agent through actions, feedback, and rewards over time, as in game playing or robotics. A fixed archive of tagged claims is not that kind of environment.
Option C mistakes the goal. The manager wants to classify existing claims, not generate new content. Writing synthetic claims does not produce a fraud score.
Reference: What is supervised learning?