Criticality prediction models using SDL metrics set

Abstract

This paper focuses on the experiences gained from defining design metrics f o r SDL and comparing three prediction models f o r identifying the most fault-prone entities using the defined metrics. Three sets of design complexity metrics f o r SDL are defined according to two design phases and SDL entity types. Two neural net based predicfion models and a model using the hybrid metrics are implemented and compared by a simulation. Though the backpropagation model shows the best prediction results, the selection method in hybrid complexity order is expected to have similar performance with some supports. Also two hybrid metric forms (weighted sum and weighted multiplication) are compared and it is shown that two metric forms can be used interchangeably f o r ordinal purpose.