City Research Online

Abstract

The aim of this thesis is to identify statistical techniques which are appropriate for the analysis of software development metrics and to investigate how they might be useful to support quality management procedures.

The initial approach was to investigate the use of statistical techniques to identify consistent relationships between measures collected during the development and fault or change-proneness of the final product. No common relationships were identified between the datasets when module relationships were considered. Therefore, there is little hope of identifying any general relationships between module attributes and product quality attributes.

However, some techniques were good at identifying outlier/anomalous components irrespective of the particular attributes. For univariate outlier detection a modification of the boxplot technique was found to be useful. This is described in the document. For bi-variate outliers, scatterplots were found to be useful. This thesis describes how the scatterplot technique can be automated to objectively identify outliers. It describes a set of rules which were implemented into a prototype. The objective was to produce a technique which most consistently identified the anomalies that had been identified subjectively by an expert consultant.

The thesis describes how summary statistics can be useful at the project level. It identifies a sub-set of useful information to enable a project manager to control his/her project. A target value, where appropriate is suggested for each measurement. Monitoring is based on the principle that when an actual attribute value exceeds the target value then it is likely to be a potential problem in the development.

A survey highlighted that for automatic anomaly detection to be of any significant benefit to a project manager, some interpretation is required to identify the likely cause of the anomaly and its effect on the project. The thesis shows how the cause of an anomaly can be diagnosed with the help of a simple expert system which looks at a combination of attribute values for diagnosis.
page

Publication Type:	Thesis (Doctoral)
Subjects:	H Social Sciences > HA Statistics Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Science & Technology > Computer Science School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads