New Software Risk Model, Using Space Shuttle Defect and Risk Data

norman schneidewind

The 17th IEEE International Symposium on Software Reliability Engineering: Fast Abstracts (ISSRE 2006)
Raleigh, North Carolina, USA, November 6-10 2006

---

Abstract

Objective of Research Our objective is to develop and demonstrate a new model for predicting risk (i.e., probability that defect counts exist on a release), as a function of risk variables. To develop and evaluate this model, we use defect and risk data from the NASA Space Shuttle Flight Software. We build upon our previous research [1] to develop predictors of risk (e.g., discrepancy report counts on a release), as a function of risk variables (sloc and issues). These predictors are significant because, with them, the risk to reliability of making requirements changes can be predicted early, when the cost and effort of correcting problems is low.

We conceptualize the model based on both previous research [1] and on new concepts. With respect to [1], we found that the risk variables sloc and issues (see the definitions below) were statistically significant predictors of cumulative failures for the Shuttle software. However, because these results were based on failures rather than defects, we would have to wait until the test phase to collect failure data to develop and evaluate the model. It is better to use defects as the dependent variable because 1) they can be collected and analyzed earlier in the development cycle and 2) the resultant model can be used earlier to identify problems when the cost of correction is lower. The new concepts are embodied in the Desirable Properties of Model below and in equations (1.1) and (1.2). We show the outlines of model development and the results of the prediction experiments.

---

  

START Conference Manager (V2.52.6)

Maintainer: mark.sherriff@ncsu.edu