ORIGINAL COMPANY-CONTEXT PRACTICE
Distinguish missing sensor data from quiet equipment
Original practice scenario: For a hypothetical energy-asset monitoring dashboard, sensor messages arrive late and some devices stop reporting. Design features that distinguish a calm process from missing evidence.
Written for independent practice using company context; this is not a verified question from a company interview.
A REASONED APPROACH
Model answer.
Keep observation value, observation time and data availability separate. A flat line constructed by forward filling is not evidence that equipment remained unchanged.
Work through it
- Retain source timestamps, arrival timestamps, sequence numbers and quality flags; measure lateness and the duration since the last valid observation.
- Define bounded resampling and imputation rules by sensor meaning, adding explicit missingness and stale-data indicators to features and displays.
- Evaluate alerts against labeled historical or synthetic sequences containing outages, drift and late backfills, using event-time windows that reproduce online visibility.
Tradeoffs to discuss
- Imputation can stabilize downstream models but obscures outages; dropping incomplete windows avoids invented values while reducing coverage.
Common pitfalls
- Do not use a late measurement in a historical prediction made before it arrived.
- Track sensor replacement and calibration changes so a device reset is not mistaken for a process anomaly.