Predicting for MTBF Failure Data Series of Software Reliability by Genetic Programming Algorithm

At present, most of software reliability models have to build on certain presuppositions about software fault process, which also brings on the incongruence of software reliability models application. To solve these problems and cast off traditional models´ multi-subjective assumptions, this paper adopts genetic programming (GP) evolution algorithm to establishing software reliability model based on mean time between failures´ (MTBF) time series. The evolution model of GP is then analyzed and appraised according to five characteristic criteria for some common-used software testing cases. Meanwhile, we also select some traditional probability models and the neural network model to compare with the new GP model separately. The result testifies that the new model evolved by GP has the higher prediction precision and better applicability, which can improve the applicable inconsistency of software reliability modeling to some extent