A new boosting-based software reliability growth model

Abstract

A new software reliability growth model (SRGM) called RBoostSRGM is proposed in this paper. It can be regarded as a modification of the boosting SRGMs through the use of a reduced set of weights to take into account the behavior of the software reliability during the debugging process and to avoid overfitting. The main idea underlying the proposed model is to take into account that training data at the end of the debugging process may be more important than data from the beginning of the process. This is modeled by taking a set of weights which are assigned to the elements of training data, i.e., to the series of times to software failures. The second important idea is that this large set is restricted by the imprecise ε-contaminated model. The obtained RBoostSRGM is a parametric model because it is tuned in accordance with the contamination parameter ε. As a variation to this model, we also consider the use of the Kolmogorov-Smirnov bounds for the restriction of the set of weights. Various numerical experiments with data sets from the literature illustrate the proposed model and compare it with the standard non parametric SRGM and the standard boosting SRGM.

Keywords:

• Software reliability growth model
• boosting
• support vector regression
• extreme points
• imprecise contaminated model
• Kolmogorov-Smirnov bounds
• pairwise comparisons

Acknowledgment

The presentation of this paper has been improved by valuable suggestions from an associate editor and a reviewer, for which we are grateful. Part of the research reported in this paper took place during a visit of Prof. Utkin to Durham University in November 2016, funded through a “Research in Pairs” (Scheme 4) grant from the London Mathematical Society.