Reliability Growth Modeling for Software Fault Detection Using Particle Swarm Optimization

Modeling the software testing process to obtain the predicted faults (failures) depends mainly on representing the relationship between execution time (or calendar time) and the failure count or accumulated faults. A number of unknown function parameters such as the mean failure function mu(t;beta) and the failure intensity function lambda(t;beta) are estimated using either least-square or maximum likelihood estimation techniques. Unfortunately, the model parameters are normally in nonlinear relationships. This makes traditional parameter estimation techniques suffer many problems in finding the optimal parameters to tune the model for a better prediction. In this paper, we explore our preliminary idea in using particle swarm optimization (PSO) technique to help in solving the reliability growth modeling problem. The proposed approach will be used to estimate the parameters of the well known reliability growth models such as the exponential model, power model and S-shaped models. The results are promising.