Advanced models for software reliability prediction

This article describes the advanced parametric models for assessment and prediction of software reliability, based on statistics of bugs at the initial stage of testing. The parametric model approach, commonly associated with reliability issues, deals with the evaluation of the amount of bugs in the code. Computed parameter values inserted into the model allow to estimate: (a) number of bugs remaining in the product, and (b) time required to detect the remaining bugs. Many models are developed for similar purpose: Duane Reliability Growth Model, Goel Model, Weibull Model, Classical S-shaped Model, Ohba S-shaped Model, etc. Taking into account some detailed, but practical, aspects of the software testing process, a few Advanced Models were developed and usefully implemented by the authors. The proposed models are sensitive to the situations typical for the early stages of Software development. As a result, one deals with the essentially non-linear, multimodal goal function to define the optimal value as the estimation of the unknown control parameter. To support the optimization of such complex models, the Cross-Entropy Global Optimization Method is proposed. Some authentic numerical examples are considered to demonstrate the efficiency of the proposed models.