(An) input domain selection method based on multi-criteria in data-driven software reliability model

Abstract

An analytical software reliability model that predicts future software failures based on software failures during the test phase has been studied for decades. However, the analytical software reliability model has limitations such as unrealistic assumptions and model selection problems, and a data-driven software reliability model has been proposed to overcome them. The data-driven software reliability model generally has a Multiple Delayed Input Single Output structure in which the cumulative failures from t-1 to t-d are input in order to predict the cumulative failures at time t, where d is the size of the input domain. Previous studies have shown that the data-driven software reliability models with different input domains have gap of predictive performance, but the research assumed that there is a method to select the input domain with the best predictive performance. However, in practice, we cannot use data driven software reliability model with an optimal input domain selection method, and just use data driven software reliability model with a heuristic input domain selection method. In this paper, we propose an input domain selection method based on multi-criteria in data-driven software reliability model. Using the failure datasets collected from 36 projects, we show that data-driven software reliability model using proposed selection method has better predictive performance than the existing model.