An effective parameter estimation technique for software reliability growth model using real-valued genetic algorithm

Abstract

As the complexity and size of software continues to increase, the interest in software quality is also growing. There are lots of software quality attributes, such as functionality, reliability and etc. Among these attributes, software reliability is considered as an important factor in software quality, since it is possible to quantify software failures. To evaluate software reliability, Software Reliability Growth Model (SRGM) has been widely used in many industrial fields. SRGM estimates software reliability, based on the detected failure data during a testing phase. By using SRGM, project managers can determine the release time and manage the testing efforts in their project environments. Traditional SRGM has a mean value function which represents the number of failures at the current time, and this function has several parameters determining the accuracy of each model. In general, numerical techniques, such as Maximum Likelihood Estimation (MLE) and Least Square Estimation (LSE), are used to estimate the parameters of SRGM. However, these techniques are useful only in linear functions or simple datasets. If a modeling function is non-linear function or the size of data is large, these techniques cannot solve a local optimization problem. Since the mean value function of SRGM is generally a non-linear function, these techniques are not suitable for the parameter estimation of SRGM. Therefore, Meta-heuristic algorithms, such as Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), have been recently used for estimating parameters of SRGM. In this paper, we propose an effective parameter estimation technique of SRGM using Real-valued Genetic Algorithm (RGA). The proposed technique applies real-valued approach to improve the accuracy and performance of the parameter estimation of SRGM. We conducted experiments on six datasets for comparing the accuracy and stability of the proposed technique with numerical techniques and existing GA techniques. The re...