A robust weighted SVR-based software reliability growth model

Abstract

This paper proposes a new software reliability growth model (SRGM), which can be regarded as an extension of the non-parametric SRGMs using support vector regression to predict probability measures of time to software failure. The first novelty underlying the proposed model is the use of a set of weights instead of precise weights as done in the established non-parametric SRGMs, and to minimize the expected risk in the framework of robust decision making. The second novelty is the use of the intersection of two specific sets of weights, produced by the imprecise ε-contaminated model and by pairwise comparisons, respectively. The sets are chosen in accordance to intuitive conceptions concerning the software reliability behaviour during a debugging process. The proposed model is illustrated using several real data sets and it is compared to the standard non-parametric SRGM.