异常检测
分离(微生物学)
计算机科学
异常(物理)
概念漂移
数据挖掘
特征(语言学)
人工智能
数据流挖掘
语言学
哲学
物理
微生物学
生物
凝聚态物理
作者
Nidhi Ahlawat,Amit Awekar
标识
DOI:10.1145/3632410.3632486
摘要
Isolation Forest (iForest) is a well-known model for anomaly detection task. It works by identifying regions corresponding to existing anomalies in the data. With the arrival of new data, concept drift can occur in two ways. First, anomalies can occur in the new regions of the feature space. Second, existing anomalies can become normal with the addition of new data. We observe that the performance of Isolation Forest severely degrades in both these scenarios. Current works fail to tune the existing Isolation Forest to adapt to the concept drift. We propose Incremental Isolation Forest to quickly update the existing Isolation Forest in response to the arrival of new data. Initial experimental results using three real-world datasets indicate that our approach achieves significant time savings with minimal loss in anomaly detection performance.
科研通智能强力驱动
Strongly Powered by AbleSci AI