Incremental Learning with Concept Drift Detection and Prototype-based Embeddings for Graph Stream Classification

TL;DR

The paper tackles graph stream classification in nonstationary environments where graphs vary in size and structure over time. It introduces a hybrid incremental learning framework that uses per-class prototypes to form graph embeddings and a loss-based concept drift detector to trigger prototype re-calculation, thereby maintaining accurate predictions as the data drift. Prototypes are selected via the Centers algorithm from class-specific memory and used to embed graphs by distances to prototypes, with incremental training adapting the classifier based on streaming embeddings. Experimental results on Letter, GREC, and Fingerprint datasets show that the embedding-based approach outperforms a hand-crafted feature baseline, and that the drift detector enhances adaptation by updating prototypes after drift events, highlighting practical value for evolving graph domains like critical infrastructure and social networks. The work offers a scalable, online solution for robust graph classification in nonstationary streams with minimal retraining, and points to future directions in learning embeddings directly and extending to unsupervised or few-shot settings.

Abstract

Data stream mining aims at extracting meaningful knowledge from continually evolving data streams, addressing the challenges posed by nonstationary environments, particularly, concept drift which refers to a change in the underlying data distribution over time. Graph structures offer a powerful modelling tool to represent complex systems, such as, critical infrastructure systems and social networks. Learning from graph streams becomes a necessity to understand the dynamics of graph structures and to facilitate informed decision-making. This work introduces a novel method for graph stream classification which operates under the general setting where a data generating process produces graphs with varying nodes and edges over time. The method uses incremental learning for continual model adaptation, selecting representative graphs (prototypes) for each class, and creating graph embeddings. Additionally, it incorporates a loss-based concept drift detection mechanism to recalculate graph prototypes when drift is detected.

Figures (9)

• Figure 1: Overview of the proposed method.
• Figure 2: Class prototypes from graph memory.
• Figure 3: Distortions levels of character "A" in the Letter dataset.
• Figure 4: Distortion levels of GREC images.
• Figure 5: Examples of the two classes in the Fingerprint dataset.