Q-YIELD™ FAQ

How does Q-Yield Work?

Q-YIELD is a nugget finder.

Given a specification of a set of records (the problem definition or target problem) Q-YIELD performs a fast heuristic search for an expression which best describes that set or a reasonable subset of it. The expressions found will typically be of the form:

<variable> <relational operator> <constant> AND <variable> <relational operator> <constant> AND ...

(although this is not always the case.)

In deriving these expressions it uses basic statistical tests and thresholds to reject terms in an expression which are not statistically significant, and to reject expressions or rules which do not describe a significant subset of the target set. Thus if we set as a problem a description of the set of records which contain abnormal process performance and provide values such as process parameters and equipment identifiers which fully describe key process events, we will quickly search through the space of possible explanations of the abnormality.

In contrast to many other approaches, the heuristics used by Q-YIELD do not make any assumptions about the distribution of data values. Q-YIELD therefore continues to perform well in the presence of highly skewed distributions, or distributions with outliers. The evaluation function used is also relatively insensitive to noise.

Once such a rule or nugget has been found, records in the subset defined by the rule are excluded from the data set, and a search is made for further nuggets. Thus, for example, if a problem has multiple causes then all cases explainable by the first rule will be excluded from the data, and the search will continue with the remaining data for possible secondary effects.

Back to FAQ