Time series one class classification: a systematic review of methods, applications, and challenges

Time series one-class classification (TS-OCC) has become an important machine learning paradigm for detecting anomalies and verifying patterns in domains where abnormal data are scarce or difficult to label, such as precision manufacturing, healthcare monitoring, finance, and cybersecurity. Unlike traditional multi-class classification, TS-OCC models are trained exclusively on normal data and identify deviations that indicate faults, failures, or irregular events. This review explores the evolution of TS-OCC methods for modelling, detecting, and interpreting anomalies in temporal data. Using a systematic approach, the review surveys literature from Scopus, and Web of Science, classifying existing methods under six categories: distance-based, boundary-based, density-based, reconstruction-based, feature-representation, and contrastive-representation approaches. These methodological paradigms are critically compared with respect to their modelling principles, representative algorithms, and benchmark datasets, as well as evaluation strategies, computational characteristics, and application domains. It also explores existing problems such as threshold calibration, concept drift, explainability, benchmarking, computational efficiency and deployment in resource-constrained environments. This review aims to bridge the gap between conceptual understanding and practical deployment insights, thus offering a roadmap to researchers, engineers, and practitioners working in this fast-evolving domain.