Anomaly detection is crucial across multiple domains like finance, security, and manufacturing. Isolation-based methods offer advantages such as low complexity, scalability, and robustness. They rely on partitioning data to isolate anomalies efficiently. Various strategies like axis-parallel splitting, random hyperplanes, hyperspheres, Voronoi diagrams, and hash-based splitting are used for isolation. The path length and hypersphere size play a key role in determining anomaly scores. Isolation mechanisms have been extended to detect group anomalies using the Isolation Distributional Kernel (IDK). Applications include detecting anomalies in streaming data, time series, trajectory datasets, images, texts, and more. Parameter optimization and model optimization are essential for improving the performance of isolation-based algorithms.
他の言語に翻訳
原文コンテンツから
arxiv.org
深掘り質問