Software Reliability Estimation with ART Network of Artificial Neural Network Using Execution Time Model

Abstract

For estimating the software reliability, it is required to observe its failure intensity. As failure intensity depends upon the number of faults, so to find the number of faults we are using the adaptive resonance theory (ART) of ANN, which is based on the best match strategy of competitive learning. The ART is able to incorporate the two different modes i.e. plasticity and stability [1]. This method provides the direct mapping between existing similarities so that the networks find the sufficiently closed match with the input pattern and the corresponding number of faults can be estimated. If the unknown prototype input pattern belongs to any generated category of the network then network displays the accretive behavior. In this case the corresponding number of faults will be same as the already defined number of faults for that group through the predictive unit. If the presented prototype input pattern does not belong to any generated category of the network that the network shows the interpolative behavior, the corresponding faults for this prototype input pattern can be determined from the average of the faults in the neighboring groups of already trained pattern. This new group will be neighbor of all the groups that shows the approximate same orientation.