A measure to distinguish between a logistic curve model and a Gompertz curve model
Daisuke Satoh · 2003
Many software reliability growth models have been proposed and analyzed for measuring the growth in software reliability. However, far fewer comparison criteria have been proposed than the number of reliability models. We propose a measure to determine whether the logistic curve model or Gompertz curve model is more appropriate for analyzing a data set. The two models are the simplest models for estimating an S-shaped software reliability growth process, which is often observed in actual projects. The two models used in the proposed measure are discrete ones. One is based on a discrete equation proposed by Morishita and the other is based on a discrete equation that is proposed in this paper. These discrete equations have exact solutions. These two models reproduce the values of the parameters perfectly when exact solutions are used as an input data. Estimated parameters are independent of time scale. The two models enable us to accurately estimate parameters in the early testing phase with actual data. I. INTRODUCTION Software reliability assessment is a key technology for re- ducing software costs and producing highly reliable software. Many software reliability growth models have been proposed and analyzed for measuring software reliability growth. Re- cently, discrete analogs of software reliability growth models have been proposed (3), (4). These provide accurate parameter estimates using the data available during the early testing phase. However, software reliability growth models that only yield accurate parameter estimates in the early testing phase do not fulfill the requirements of software engineers and managers because such models do not provide grounds for selecting the most appropriate model. Far fewer comparison criteria have been proposed than the number of reliability models. In this paper, we propose a measure to distinguish between a logistic curve model and a Gompertz curve model, which are simple S-shaped models. We also propose a new discrete Gompertz curve model to use it in the measure.