Concept Drift Detection for Streaming Data Using One-Class Classification

Author

• Poorna Sandamini Senaratne, University of Central FloridaFollow

ORCID

0009-0009-2008-2836

Keywords

One-Class Classification, Embeddings, LS-SVDD, Concept Drift, Streaming Data, Neural Networks

Subject Categories

Artificial Intelligence and Robotics | Computer Sciences

Abstract

Modern machine learning systems are increasingly deployed in streaming environments where data arrive sequentially and the underlying data-generating process may evolve over time. This phenomenon, known as concept drift, can significantly degrade model performance if not detected and addressed in a timely manner. This dissertation proposes a principled framework for concept drift detection based on one-class classification, integrating neural network embeddings with Support Vector methodologies.

The proposed approach leverages neural networks to learn compact and informative embeddings of input data, capturing complex nonlinear structures in a lower-dimensional latent space. These embeddings are then used to construct a statistical description of normal behavior via Support Vector methods. Concept drift is detected by monitoring deviations in the distance of incoming observations from the learned support vector data description boundary.

The effectiveness of the proposed framework is evaluated on both synthetic and real-world streaming datasets, including the Internet Firewall, SINE1 dataset and the ELEC2 electricity market dataset. Experimental results demonstrate that the method is capable of detecting both gradual and abrupt distributional changes, often identifying drift earlier than established benchmark methods. Additionally, the framework exhibits resilience to noise while maintaining sensitivity to meaningful shifts in the data distribution.

Overall, this work provides a flexible and interpretable approach to concept drift detection, combining deep representation learning with statistically grounded one-class modeling. The proposed methodology offers a strong foundation for developing adaptive and reliable machine learning systems in dynamic, real-world environments.

Completion Date

2026

Semester

Summer

Committee Chair

Maboudou, Edgard

Degree

Doctor of Philosophy (Ph.D.)

College

College of Sciences

Department

School of Data, Mathematical, and Statistical Sciences

Format

PDF

Document Type

Dissertation

Language

English

STARS Citation

Senaratne, Poorna Sandamini, "Concept Drift Detection for Streaming Data Using One-Class Classification" (2026). Graduate Studies Theses and Dissertations 2026. 349.
https://stars.library.ucf.edu/gradstudies_etd_2026/349