ENHANCING SOFTWARE RELIABILITY PREDICTION USING NN, GP, ACOT, AND PSO

Abstract

Software reliability prediction is required to build software systems of good quality. In this paper, the performance of evolutionary computing models, Neural Networks (NN), Genetic Programming (GP), Ant Colony Optimization Technique (ACOT), and Particle Swarm Optimization (PSO), for software reliability prediction is compared. Randomness and variance of software behaviors do not make it possible for conventional methods to estimate its reliability. There is enough evidence that evolutionary models such as Neural Networks (NN), Genetic Programming (GP), Ant Colony Optimization Technique (ACOT), and Particle Swarm Optimization (PSO) can be utilized to solve the issues. NN, GP, ACOT, and single PSO are single models under consideration in this paper for possible utilization to software reliability prediction. The performance of the models is compared with a benchmark dataset. Each of the models is compared according to prediction accuracy, convergence rate, and autocorrelation values. To observe how well they are performing, their accuracy, convergence rate, and autocorrelation effect is examined. From the outcomes, although all four models are good predictors, NN and PSO are better in terms of convergence rate and accuracy. Hybrid systems utilizing multiple techniques provide a better accuracy-computational cost trade-off from the outcomes. This suggests that model combination is a good area to improve the performance of software reliability prediction.