Machine Learning Under Concept Drift in Highly Dynamic Environments: Advances and Challenges (MLCD-2025)

Abstract

The increasing deployment of machine learning systems into real-world applications that are subsumed within constantly changing environments has brought renewed attention to concept drift—the change in the underlying data distribution of the problem being modelled over time.

Examples of such environments range from predictive maintenance and financial modelling to health monitoring and climate science. In those highly evolving data streams, machine learning models are significantly challenged in their performance, and therefore, demand specific approaches to incorporate new capabilities for adapting to change into classical machine learning algorithms and strategies. Approaches that must deal with drift detection, characterisation, and action, either in a preventive or reactive fashion, in a way that ML-based systems can continue to perform their tasks with acceptable performance. White-box and black-box approaches for the development of drift-aware ML model engineering will be considered, along with all phases of model development, including data gathering and preparation, model selection, validation, deployment, and interpretation.

This special session aims to gather researchers and practitioners working on theoretical foundations, practical algorithms, and real-world applications that address the theme of concept drift. Topics of interest include, but are not limited to:

Topics

● Drift detection and adaptation techniques
● Drift-aware machine learning techniques and models
● Evaluation metrics and benchmark datasets
● Explainability and uncertainty under drift
● Domain Adaptation
● Data shifts and data quality with regard to drift analysis
● Hybrid and ensemble methods for non-stationary data
● Online learning and incremental algorithms
● Applications in healthcare, finance, cybersecurity, IoT, and more

The session, an activity associated with the CALM project, will serve as a platform to share recent advances, foster interdisciplinary collaboration, and stimulate discussion on future directions in concept drift.

Organizers

Dr. José Tomás Palma Méndez, Department of Information and Communications Engineering, University of Murcia (jtpalma@um.es)
Dr. Juan A. Botía Department of Information and Communications Engineering, University of Murcia
Grzegorz J. Nalepa, Halmstad University, ELLIIT, Sweden and Jagiellonian University, Krakow, Poland
Martin Atzmueller, Institute of Computer Science at Osnabrück University and German Research Center for Artificial Intelligence (DFKI)

Submission

See submission instructions for the conference at the call for papers. At the beginning of the submission, please choose the track “Machine Learning Under Concept Drift in Highly Dynamic Environments: Advances and Challenges”.

Submission link: https://easychair.org/conferences/?conf=ideal2025

Special Session Papers Submission Deadline: June 15th, 2025