MLA-C01 Practice Question 34: Unsupervised Anomaly Detection
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 4 · question 4 of 10
A utility collects per-minute readings from thousands of smart meters, with about 20 numeric features per reading. It wants to flag unusual readings, such as sudden spikes or broken daily patterns, that may point to faulty meters or tampering. There are no labels: nobody has marked which historical readings were abnormal. The team wants a SageMaker AI built-in algorithm that assigns each reading a score indicating how anomalous it is.
Which built-in algorithm should the team use?
Show answer and explanation
Answer:
Random Cut Forest (RCF) is SageMaker AI's unsupervised algorithm for detecting anomalous data points. It needs no labels, works on inputs with any number of dimensions, and returns an anomaly score for every data point: low scores are normal, high scores are anomalous. A common rule of thumb treats scores more than three standard deviations above the mean as anomalies (option C).
- A is wrong because XGBoost as a classifier is supervised. It needs examples labeled normal and anomalous, and the scenario says none exist. - B is wrong because Linear Learner predicts a target. Turning its prediction errors into an anomaly detector is a custom design built around the wrong tool, and it does not give the ready-made anomaly score the team asked for. - D is wrong because DeepAR forecasts time series. Comparing forecasts with actual readings can reveal anomalies, but that is extra engineering on top of a forecaster. It is not a purpose-built anomaly scorer for multi-feature readings.
Reference: Random Cut Forest (RCF) Algorithm