An aircraft engine manufacturer collects 200 time-series metrics during testing and wants near-real-time detection of critical defects while retaining all data for offline analysis. Which approach is MOST effective for near-real-time anomaly detection?

The correct answer leverages Amazon Kinesis Data Firehose for efficient ingestion and delivery of streaming data, and Amazon Kinesis Data Analytics with a Random Cut Forest (RCF) application for near-real-time anomaly detection. RCF is well-suited for time-series anomaly detection. Firehose also persists the data to Amazon S3, fulfilling the requirement for offline analysis. Incorrect options: AWS IoT Analytics is suitable for IoT data, but running Jupyter notebooks for near-real-time detection is not its primary strength for continuous streaming anomaly detection. Using Amazon EMR with Apache Spark and k-means clustering can detect anomalies, but it's typically more batch-oriented and might not meet the "near-real-time" requirement as effectively as Kinesis Data Analytics. Using Amazon SageMaker's RCF algorithm offline would not provide near-real-time detection; it's designed for batch processing of pre-collected data.