The Functional Isolation Forest (FIF) algorithm, while effective for anomaly detection in functional data, suffers from limitations related to its reliance on a predefined dictionary and linear inner product for scoring abnormalities.
This paper presents the Signature Isolation Forest, which utilizes the signature method from rough path theory to enhance anomaly detection algorithms by eliminating restrictive dependencies of FIF.
The proposed K-SIF and SIF algorithms significantly outperform the existing Functional Isolation Forest in handling complex datasets and provide a more robust framework for real-data anomaly detection based on comprehensive numerical experiments.
Through extensive experiments, our results validate the advantages of the Signature Isolation Forest approach over traditional methods, demonstrating its effectiveness in identifying anomalies in practical applications.
Collection
[
|
...
]