Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Applied Mathematics

Supervisor

Matt Davison

Abstract

The field of Software Engineering (SE) is the study of systematic and quantifiable approaches to software development, operation, and maintenance. This thesis presents a set of scalable and easily implemented techniques for quantifying and mitigating risks associated with the SE process. The thesis comprises six papers corresponding to SE knowledge areas such as software requirements, testing, and management. The techniques for risk management are drawn from stochastic modeling and operational research.

The first two papers relate to software testing and maintenance. The first paper describes and validates novel iterative-unfolding technique for filtering a set of execution traces relevant to a specific task. The second paper analyzes and validates the applicability of some entropy measures to the trace classification described in the previous paper. The techniques in these two papers can speed up problem determination of defects encountered by customers, leading to improved organizational response and thus increased customer satisfaction and to easing of resource constraints.

The third and fourth papers are applicable to maintenance, overall software quality and SE management. The third paper uses Extreme Value Theory and Queuing Theory tools to derive and validate metrics based on defect rediscovery data. The metrics can aid the allocation of resources to service and maintenance teams, highlight gaps in quality assurance processes, and help assess the risk of using a given software product. The fourth paper characterizes and validates a technique for automatic selection and prioritization of a minimal set of customers for profiling. The minimal set is obtained using Binary Integer Programming and prioritized using a greedy heuristic. Profiling the resulting customer set leads to enhanced comprehension of user behaviour, leading to improved test specifications and clearer quality assurance policies, hence reducing risks associated with unsatisfactory product quality.

The fifth and sixth papers pertain to software requirements. The fifth paper both models the relation between requirements and their underlying assumptions and measures the risk associated with failure of the assumptions using Boolean networks and stochastic modeling. The sixth paper models the risk associated with injection of requirements late in development cycle with the help of stochastic processes.

Share

COinS