Rough set-inspired isolation forest

This paper presents an innovative approach to anomaly detection, combining the Isolation Forest method with Zdzisław Pawlak's rough set theory. The core methodology involves modifying the structure of binary trees used in Isolation Forests, allowing nodes to have more than the usual two childre...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Information sciences Jg. 718; S. 122390
Hauptverfasser: Rachwał, Albert, Karczmarek, Paweł, Rachwał, Alicja
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Inc 01.11.2025
Schlagworte:
Anomaly detection
Binary trees
Isolation forest
Pattern recognition
Rough set theory
Tree algorithms
Tree algorithms
Rough set theory
Isolation forest
Pattern recognition
Anomaly detection
Binary trees
ISSN:0020-0255
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This paper presents an innovative approach to anomaly detection, combining the Isolation Forest method with Zdzisław Pawlak's rough set theory. The core methodology involves modifying the structure of binary trees used in Isolation Forests, allowing nodes to have more than the usual two children. Based on rough set theory, two approaches are developed: one where each node has three children, and another with five children per node. According to Pawlak's theory, lower and upper approximations are sought using the mean and standard deviation to define rough sets. This methodology is further enhanced by introducing a boundary region in the partitioned dataset, creating two variants: one with boundaries spanning from 1.5 to 3 standard deviations and another from 3 to 6 standard deviations. Each attribute is divided into a rough set comprising five subsets, with observations assigned to specific areas based on the defined limits, receiving corresponding weights. The proposed models are evaluated against the base Isolation Forest and K-Means-Based Isolation Forest, demonstrating notable improvements in performance. •Tree nodes are enhanced with rough set-based approximations.•Isolation Forest and its modifications are compared for efficiency.•Modifications significantly improve anomaly detection scores.
ISSN:0020-0255
DOI:10.1016/j.ins.2025.122390