Using Feature Store online and offline stores for fraud scoring

MLA-C02 Data Preparation for ML and AI Medium Question 8 of 30

A financial services company is developing a real-time fraud detection platform on AWS. An Amazon SageMaker AI real-time endpoint must retrieve current customer feature values with single-digit millisecond latency during transaction scoring, while the data science team requires historical feature values spanning several years to build training datasets. An ML engineer creates an Amazon SageMaker Feature Store feature group configured with both an online store and an offline store.

Which solution will meet these requirements?

Show answer and explanation

BUse the online store GetRecord API for real-time inference and query the offline store with Amazon Athena for training.

Using the online store with the GetRecord API delivers single-digit millisecond latency for real-time fraud scoring because the online store retains the latest feature values in an optimized key-value cache, while querying the offline store with Amazon Athena allows the data science team to extract historical records stored in Amazon S3 for model training. Reversing the access paths is incorrect because Athena queries introduce seconds of latency during inference, and the online store cannot provide historical state. Relying solely on Athena queries on the offline store is incorrect because Athena query execution exceeds the single-digit millisecond latency required for online scoring. Relying solely on the online store for training is incorrect because the online store retains only the latest records.

Source: docs.aws.amazon.com/sagemaker/latest/dg/feature-store.html

Anonymous, no account needed. Your tally stays in this browser.

Next question Start 5 timed questions Browse all 30 original MLA-C02 practice questions