A novel machine learning approach for software reliability growth modelling with pareto distribution function

Software reliability is the important quantifiable attribute in gaining reliability by assessing faults at the time of testing in the software products. Time-based software reliability models used to identify the defects in the product, and it is not suitable for dynamic situations. Instead of time, test effect is used in few explorations through effort function and it is not realistic for infinite testing time. Identifying number of defects is essential in software reliability models, and this research work presents a Pareto distribution (PD) to predict the fault distribution of software under homogenous and nonhomogeneous conditions along with artificial neural network (ANN). This methodology enables the parallel evolution of a product through NN models which exhibit estimated Pareto optimality with respect to multiple error measures. The proposed PD-ANN-based SRGM describes types of failure data and also improves the accuracy of parameter estimation more than existing growth models such as homogeneous poison process and two fuzzy time series-based software reliability models. Experimental evidence is presented for general application and the proposed framework by generating solutions for different product and developer indexes.