Doctoral Dissertations

Keywords and Phrases

Cluster Validity; Clustering; Continuous Learning; Explainable AI; Lifelong Learning; Supervised Learning

Abstract

Clustering and supervised learning are often treated as distinct paradigms, yet both rely on structure in feature space. This dissertation investigates the relationship between cluster validity indices (CVIs) and supervised learning in real-time and lifelong learning settings where data arrive incrementally and cannot be revisited. Across four studies, it develops methods for online cluster validation, uses supervised learning to improve their interpretability, and applies these ideas to evaluating performance degradation in continual learning.

The first study extends incremental cluster validity indices (iCVIs), enabling widely used validation metrics to operate in streaming environments. Experiments on synthetic and real-world datasets show systematic differences in how iCVIs respond to under- and over-partitioning, highlighting both their usefulness and their limitations. The second study introduces the Meta iCVI, an ensemble framework that uses supervised learning to combine multiple iCVIs into a single interpretable measure of partition quality, improving the labeling of clustering outcomes.

The third study examines supervised lifelong learning and shows that performance degradation cannot always be attributed to catastrophic forgetting. It identifies class overlap in feature space as an independent source of degradation, termed memory overshadowing, which imposes a fundamental performance bound regardless of model memory. The fourth study builds on this result by introducing new iCVI-derived metrics for supervised evaluation: the Overlap Index, which measures feature-space overlap, and the Overshadowing and Forgetting Index, which separates true forgetting from data-induced interference.

Together, these contributions connect cluster validation and supervised learning evaluation, providing both theoretical insight and practical tools for interpreting model performance in incremental, non-stationary settings.

Advisor(s)

Wunsch, Donald C.

Committee Member(s)

Yang, Huiyuan
Chatterjee, Shubham
Corns, Steven
Maity, Suman

Department(s)

Electrical and Computer Engineering

Second Department

Computer Science

Degree Name

Ph. D. in Computer Science

Publisher

Missouri University of Science and Technology

Publication Date

Spring 2026

Journal article titles appearing in thesis/dissertation

Paper I: Pages 4-72 have been published in IEEE Access, vol. 8, 2020.

Paper II: Pages 73-98 have been published in IEEE Access, vol. 12, 2024.

Paper III: Pages 99-115 have been published in the proceedings of the 2025 12th International Conference on Soft Computing & Machine Intelligence (ISCMI), Rio de Janeiro, Brazil, 2025.

Paper IV: Pages 116-150 have been published in the proceedings of the 40th Annual AAAI Conference on Artificial Intelligence, Singapore, Singapore, 2026.

Pagination

xiv, 154 pages

Note about bibliography

Includes_bibliographical_references_(pages 147-150)

Rights

© 2026 Niklas Max Melton , All Rights Reserved

Document Type

Dissertation - Open Access

File Type

text

Language

English

Thesis Number

T 12600

Download

Share

 
COinS