Dissertations, Theses, and Capstone Projects

Date of Degree

9-2026

Document Type

Doctoral Dissertation

Degree Name

Doctor of Philosophy

Program

Computer Science

Advisor

Hui Chen

Committee Members

Ping Ji

Anita Raja

Kostadin Damevski

Hui Chen

Subject Categories

Artificial Intelligence and Robotics | Software Engineering

Keywords

Just-in-Time Software Defect Prediction (JIT-SDP), Software Quality Assurance, Vulnerability Patch Curation, Incremental Learning, Uncertainty Quantification, Defect types

Abstract

Software defect prediction (SDP) helps developers prioritize maintenance, reduce costs, and improve software quality. Just-In-Time SDP (JIT-SDP), which predicts defects in code changes rather than static artifacts, offers faster feedback but has reached a performance bottleneck. This dissertation systematically surveys the field, identifies key limitations, including data scarcity, label noise, concept drift, class imbalance, and the lack of uncertainty or severity information in binary predictions, and proposes four novel solutions: (1) DeepICP, an incremental learning framework that preserves temporal and structural relationships among changesets; (2) an uncertainty quantification method to curate high-quality vulnerability patch datasets; (3) a semantic feature extraction approach using large language models to enable defect category prediction beyond binary classification; and (4) a project-agnostic diagnostic framework that identifies where models succeed and fail by automatically deriving defect types from embeddings. Key findings show that performance plateaus are concentrated in specific defect types, prediction difficulty increases from type cores to boundaries, and targeted augmentation outperforms indiscriminate approaches. This work provides practical frameworks, datasets, and a foundation for advancing JIT-SDP in real-world quality assurance systems.

Download

This work is embargoed and will be available for download on Wednesday, March 31, 2027

Share

COinS