Technical debt decision-making under uncertainty

Download Statistics

Downloads

Downloads per month over past year

Fungprasertkul, Suwichak ORCID: https://orcid.org/0000-0003-0579-5781 (2024). Technical debt decision-making under uncertainty. University of Birmingham. Ph.D.

Preview

Fungprasertkul2024PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (1MB) | Preview

Abstract

The key to effective technical debt management (TDM) decision-making is to select a feasible and conclusive decision on time. Decision-making, in general, can suffer from uncertainties, obscuring a problem domain and challenging one from producing a useful decision. However, the presence of uncertainties in TDM decision-making is inadequately explored in the Technical Debt (TD) community.

This thesis investigates the presence of Information Uncertainty (uncertainty due to information deficiency), Deep Uncertainty (uncertainty due to unresolvable disagreement) and Erratic Uncertainty (uncertainty due to unpredictability of a problem domain) and their consequences in TD identification, TD measurement, TD prioritisation and TD monitoring.

We leverage goal-obstacle and the expected value of information to strategically mitigate Information Uncertainty in TD identification and measurement. By leveraging these analyses, Information Uncertainty can be managed regardless of information availability.

We leverage Many Objective Robust Decision-Making to implement a framework to mitigate Deep Uncertainty during TD prioritisation. The framework’s premise is to frame TD rankings in disagreements into different objective functions and utilise a set of decisions that perform well in those disagreements.

We propose a method to mitigate the consequence of Erratic Uncertainty in TDM monitoring: ineffective responsibility sharing. The method leverages Behavioural Cloning to filter the situation when developers/maintainers are necessary and utilises Replicator Dynamics to automatically resolve the TD monitoring situation when they are unnecessary, resulting in a more balanced workload between TD analysts and developers/maintainers.

The thesis individually evaluates the above contributions through various experiments with industrial cases, aiming for the effectiveness of the contributions compared to traditional approaches without the acknowledgement of uncertainties.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):