Evolving Machine Learning in Non-Stationary Environments: A Unified Survey of Drift, Forgetting, and Adaptation

Abstract

In an era defined by rapid data evolution, traditional Machine Learning (ML) models often struggle to adapt to dynamic environments. Evolving Machine Learning (EML) has emerged as a pivotal paradigm, enabling continuous learning and real-time adaptation to streaming data. While prior surveys have examined individual components of evolving learning - such as drift detection - there remains a lack of a unified analysis of its major challenges. This survey provides a comprehensive overview of EML, focusing on four core challenges: data drift, concept drift, catastrophic forgetting, and skewed learning. We systematically review over 100 studies, categorizing state-of-the-art methods across supervised, unsupervised, and semi-supervised learning. The survey further explores evaluation metrics, benchmark datasets, and real-world applications, offering a comparative perspective on the effectiveness and limitations of current approaches and proposing a taxonomy to organize them. In addition, we highlight the growing role of adaptive neural architectures, meta-learning, and ensemble strategies in managing evolving data complexities. By synthesizing insights from recent literature, this work not only maps the current landscape of EML but also identifies key research gaps and emerging opportunities. Our findings aim to guide researchers and practitioners in developing robust, ethical, and scalable EML systems for real-world deployment.