Why it matters
Without normalisation, the noisiest channel dominates every comparison. A z-score turns 'unusual' into a number that means the same thing on a gas channel and a pressure channel.
Prediction & analysis
Z-score
A z-score expresses how many standard deviations a value sits from the mean of its distribution, making deviations comparable across channels with different units and scales.
How it works
Subtract the channel's mean from the current value and divide by its standard deviation. Values beyond roughly two are conventionally treated as notable, though the right cut-off depends on how noisy the channel is and how costly a false alarm is.
Example
A reading three standard deviations above a node's own norm is unusual for that node regardless of the absolute figure.
In Alfa HazardX
Used to express baseline deviation, so 'unusual for this machine, in this room' is a measured quantity rather than a judgement.
Read further
Related terms
- Baseline deviationBaseline deviation measures how far a current reading has moved from what is normal for that specific sensor in that specific location.
- Deterministic analysisDeterministic analysis produces the same output for the same input every time, using explicit rules or arithmetic rather than a learned model.
- Sensor fusionSensor fusion combines readings from multiple sensing channels into a single assessment, so that patterns spanning several channels become detectable.