Time Series Anomaly Detection for Sensor Data: Models, Metrics and Methodologies – A Review

Detecting anomalies in time-series data is essential for identifying outliers and issuing early warnings of system failures in applications such as Industrial Control Systems (ICS), Internet of Things (IOT), and networks or medical risks in healthcare applications. Deep learning has greatly advanced time series anomaly detection. However, a comprehensive view of the latest architectures and their practical characteristics remains limited. This review surveys the current state of the art of 43 papers for the years 2019 to 2025, focusing on the most effective models (Transformers, Graph Neural Network (GNN), Attention Networks-based designs) and analyzes them through different metrics. Our analysis shows that many studies do not report essential metrics such as parameter count, computational cost, and predictive uncertainty, all of which govern real-time use on resource-constrained hardware. We also quantify the share of papers that use public datasets, highlight papers that released new public datasets, and papers that provide open-source code and now serve as community benchmarks. With the rapid proliferation of time series anomaly detection models, researchers can struggle to choose the framework that is best suited to their own data and constraints. This paper proposes a methodology-driven taxonomy. It classifies frameworks into six categories of methodologies: clustering, prediction, reconstruction, representation learning, statistical, and hybrid. The classification is based on the underlying methodology rather than the specific framework technology that is employed. For each methodology, we explain the core idea, examine the latest leading frameworks in this methodology, identify the most appropriate data types and use cases, indicate the commonly used implementation technologies, list typical application domains, and summarize its main strengths and limitations. We also discuss open research challenges to guide future work in the field.