Use this URL to cite or link to this record in EThOS:
Title: Analysis methods for software reliability data.
Author: McCollin, Christopher.
Awarding Body: Nottingham Trent University
Current Institution: Nottingham Trent University
Date of Award: 1993
Availability of Full Text:
Access through EThOS:
This thesis reviews the statistical models commonly applied to software reliability data. A data set encompassing the typical fields to be found on a software defect record sheet is analysed in a systematic way to initially determine where data was corrupted or uncollected. The data when summarised into failure counts, proportions, waiting times to failure and cumulative failure times are analysed by a number of statistical analyses : Exploratory Data Analysis, Box and Jenkins time series, proportional hazards modelling, proportional intensity modelling and a number of multivariate techniques. A comparison of the analyses is undertaken. The time series analysis using a standard computer package was able to forecast when the software would become failure free, a useful metric to determine time to release the software to a customer. The results are verified by proportional hazards modelling. The intensity functions of most of the non-homogeneous Poisson processes are shown to be equivalent to proportional hazards models with appropriate explanatory factors and hazard functions. The technique may be used as a diagnostic tool for the selection of the most appropriate software reliability model for a given data set as nonsignificant proportional hazards formulations are rejected from the analyses. Covariates which describe the attributes of the software, e.g. source program type, may also be incorporated in a proportional hazards formulation. The proportional intensity model is applied to the twelve least reliable program sources of Alvey data set number 3, the first analysis of this type for software data. This formulation can model all the software and hardware reliability growth models which can be expressed as Non-homogeneous poisson processes. The findings are compared with those from exploratory data analysis and proportional hazards modelling. The proportional intensity model is also shown to be a limiting form of the proportional odds model. The use of multivariate techniques such as principal components analysis, discriminant analysis and also generalised linear modelling to model software reliability data are described and the results are compared to the results of the analyses from exploratory data analysis and proportional intensity modelling.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Statistics Mathematical statistics Operations research Computer software Reliability (Engineering)