Microsoft Azure AI Fundamentals (AI-900)Describe fundamental principles of machine learning on AzureMedium
A manufacturing company uses sensors to collect data from its machinery. They want to identify unusual patterns or anomalies in the sensor data that might indicate equipment malfunction or impending failure, without having any prior examples of 'failure' data. Which machine learning approach is best suited for this task?
- AUnsupervised Anomaly Detection
- BReinforcement Learning
- CSupervised Regression
- DSupervised Classification
Show answer & explanationAnswer & explanation
Correct answer: A. Unsupervised Anomaly Detection
Since there are no prior labels for 'failure' data, supervised learning is not applicable. Unsupervised anomaly detection is designed to identify rare or unusual data points that deviate significantly from the majority of the data, which is ideal for detecting unexpected machinery malfunctions.
Why the other options are wrong
- B. Reinforcement learning is about an agent learning optimal actions through interaction, not pattern detection.
- C. Supervised regression predicts continuous values and also requires labeled data.
- D. Supervised classification requires labeled 'failure' examples for training, which are not available.
Unsupervised Anomaly Detection
A machine learning technique that identifies data points that do not conform to an expected pattern in a dataset, without requiring labeled examples of anomalies for training.
- Used when 'normal' behavior is known, but 'anomalous' behavior is not or is rare.
- Often involves models that learn the distribution of normal data.
- Outliers are flagged based on deviation from the learned normal distribution.
Memory trick: Unsupervised 'finds' the 'unseen' problems in the machine data.