Choosing and Checking a Model

Turn failure probabilities into inspection decisions

False alarms now cost more than misses

The component probabilities are already estimated. Botie's action threshold must reflect the supplied false-alarm and missed-failure costs.

BOTIE

Costs change from false alarm 1 and miss 4 to false alarm 3 and miss 2. How should the inspection cutoff move?
The inputs we are comparing

Estimated failure probabilities

Misses cost more

  1. 00.1
  2. 10.4
  3. 20.8

False alarms cost more

  1. 00.2
  2. 10.5
  3. 20.6
  4. 30.9

False-alarm cost

Misses cost more

1

False alarms cost more

3

Missed-failure cost

Misses cost more

4

False alarms cost more

2
Result for misses cost more
  1. 00
  2. 11
  3. 21
Choose an explanation to test
False alarms have zero cost in this toy policy

This check uses the pieces from “Apply the cost-derived inspection policy”.

BOTIE

Assume correct decisions cost nothing, false alarms also cost nothing, and misses cost five. Predict the decisions from the existing inclusive cost-derived threshold rule, including probability zero.
Estimated failure probabilities
  1. 00
  2. 10.01
  3. 20.4
False-alarm cost
0
Missed-failure cost
5

Use a number, text, or an array matching the requested result. For a matrix, put each row in brackets. Saving your prediction runs your current construction on these inputs and checks the earlier examples too.

Build a construction in the workspace above first. This case will test those pieces.

Which assumptions about probability quality and correct-decision costs make this threshold valid, and why would different assumptions require a new derivation?

Choose actions using credible probabilities and an explicit cost model rather than treating a default cutoff as universal.

This assumes credible probabilities, nonnegative costs with a positive sum, and zero cost for correct decisions. Different cost tables require a different derivation.

Work through the Python challenges →
Continue the courseDoes Eighty Percent Mean Eighty Percent? →
← Which Feature Mattered