A sensor has values far outside its normal range.
Decide whether they are faults or real events.
Choose two appropriate practices.
Use domain and acquisition records to determine meaning, and compare removal, clipping, transformation, or indicator policies.
Detailed explanation
Important failures may be removed.
Different treatments affect model quality and detection.
It may be a real failure or a measurement error.
Training and inference meanings diverge.
Check official documentation
Compare the following points with the official documentation.
Review Task 1.3 and the data-preprocessing guidance. Distinguish extreme measurements indicating equipment failure from unit mismatches or sensor measurement errors, then outline how removal, clipping, transformation, or indicators can affect the task.
Official references
What to confirm
Check the cause of an outlier and the prediction target, then compare the effects of treatment on model quality and anomaly detection before deleting it.
Key points
- Outlier
- Measurement error
- Domain knowledge
Notes
- Environment: Reference review of official documentation and hypothetical examples; no AWS resources or model training are required.
- The meaning of an extreme value depends on the target. If sensor failure is itself a target event, simply discarding it as bad data can remove the signal of interest.
- Fit data-dependent treatment such as clipping thresholds on training data, then apply consistent units and transformations to subsequent data.
Foundation review
Outlier
An observation far from the usual range of a dataset or process.
Measurement error
A value distortion caused by a sensor, collection, transmission, or recording problem.