# Catalogue of Artificial Intelligence Techniques

View Maths as: Images | MathML

## Certainty Factors

Keywords: MYCIN

### Categories: Expert Systems , Inference and Reasoning , Knowledge Representation

Author(s): Helen Lowe

Certainty factors are used in Production Rule Systems, such as MYCIN, to express and propagate degrees of belief. They are numbers in the range$\left[-1,+1\right]$

. $+1$ indicates certainty; $-1$ indicates impossibility. Measures of belief and disbelief are collected separately and combined by subtracting the latter from the former. Certainty factors are updated to take into account fresh evidence, the laws of combination differing from Bayesian Inference in that they are truth functional--the certainty of a formula is a unique function of the certainty of its subformulae. Approximate probabilistic interpretation of certainty factors can be given in terms of the likelihood ratio:

$L=\frac{P\left(e|h\right)}{P\left(e|¬h\right)}$

and the transformation:

$CF=\frac{\left(L-1\right)}{\left(L+1\right)}$

The certainty factor propagation method would produce coherent belief updates if no two rules emanate from the same premise.

### References:

• Heckerman, D., Probabilistic interpretations for MYCIN's certainty factors, Uncertainty in Artificial Intelligence (Kanal, L.N., and Lemmer, J.F. , eds.), North Holland, Amsterdam, 1986, pp.167--196.