Synthetic worked example
Before the model gets the data
Five invented readings show why units, missing values and validation rules come before a useful calculation.
My EdgeAi work included cleaning, validating and normalizing sensor data. This small example is newly made for the website. It uses invented temperature readings, not employer data or a claim about a real device.
For this example only, suppose an agreed valid range is 0–50 °C. We convert Fahrenheit to Celsius before applying the rule. A missing value stays flagged; an out-of-range reading is set aside for investigation rather than silently clipped or replaced.
Explain the result with its limits
Three readings remain: 20, 20 and 21 °C. Their mean is approximately 20.33 °C. One missing value and one out-of-range value remain visible as issues to investigate.
That summary is more useful than a number alone. It says what went into the calculation, what did not, and why. The rule is a teaching assumption—not a universal way to clean sensor data or evidence that the real temperature is correct.
Inspect before summarizing
Illustrative range: 0–50 °C. Unit normalization here means converting to Celsius, not scaling values into a model feature range.
Scroll sideways for the full table.
| Sample | Input | Decision | Accepted °C |
|---|---|---|---|
| A | 20 °C | Retain | 20 |
| B | 68 °F | Convert °F to °C | 20 |
| C | 21 °C | Retain | 21 |
| D | Missing | Missing; keep flagged | Flagged |
| E | 180 °C | Outside example range; investigate | Flagged |