.png){kind=icon}
返回
Virtual Node Isolation Forest: One-class isolation-based method for novelty detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2025.113296.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Most existing novelty detection approaches are computationally costly, facing difficulties to be applied for many low-cost industrial scenarios. To address this problem, in this paper we propose a low-complexity novelty detection method for low-cost industrial applications, called Virtual Node Isolation Forest. Virtual Node Isolation Forest combines the advantages of one-class novelty detection methods and unsupervised isolation-based outlier detection methods. It uses classical isolation trees to establish the boundary of normal data during the training stage, and then adds virtual nodes to isolation trees during the test stage. By adding virtual nodes, Virtual Node Isolation Forest transforms the classical static isolation trees to dynamic virtual node isolation trees, so that the generalization of isolation-based methods for one-class novelty detection can be significantly improved. Experiments over the public datasets and real-world applications show that Virtual Node Isolation Forest can detect novelties accurately. Moreover, the time and memory complexities of Virtual Node Isolation Forest are much lower than the most existing novelty detection approaches, and it can be deployed in low cost micro-controller units.
Keyword:
Isolation forest
Novelty detection
One-class
Virtual nodes
期刊
IF:
8
论文数:
5.8K
被引数:
3.5W
机构
引用论文
.png){kind=icon}
Efficient Unsupervised Parameter Estimation for One-Class Support Vector Machines一类支持向量机的高效无监督参数估计
PREAI
分享收藏
查看更多内容