Application of genetic algorithm approach on the software reliability growth models

Abstract

Reliability of a software system is the probability to operate without failure for a particular time period under certain environmental conditions. Software Reliability Models are the tools that are used to predict the Software fail probability over the time. Numerous Software Reliability Growth Models (SRGMs) were developed previously for reliability prediction of software systems by calculating their attributes (i.e. number of failures, fault detection rate etc.). None of them is proved to perform effective because each project has different characteristics. The main challenge of SRGMs is to identify the unknown model parameters. The models are generally in non-linear form, and is a big concern in finding optimal parameters with the existing techniques, Maximum likelihood Estimation (MLE) and Least Square Estimation (LSE). In this work, we propose a new model, which is based on the Genetic Algorithm (GA) approach, which is an evolutionary computation method to overcome the software reliability modeling difficulty. GA approach is already prove effective in solving the optimization problems. This fact makes it useful for solving the parameter estimation problem of SRGMs. Experiments are conducted to evaluate the new model with GA approach and compare its performance with the other models from literature that uses the traditional numerical methods (MLE and LSE) for its parameter estimation problem. Based on the findings in this thesis, the experiment results show that Genetic Algorithm (GA) generates the most accurate results as compared to the existing numerical techniques of parameter estimation problem of (SRGMs). Therefore, Genetic Algorithm is an effective approach that provides accurate and reliable parameter estimation for SRGMs.