Comparative Study of Isolation Forest and LOF algorithm in anomaly detection of data mining
Linchang Fan, Jinqiang Ma, Jun-Jing Tian, Tonghan Li, Hao Wang
Abstract
[Purpose] In the research of data mining, anomaly detection algorithms can accurately find samples of abnormal behaviors to achieve the purpose of data mining. the isolation forest algorithm and the LOF algorithm play an important role as the classic representatives of anomaly detection algorithms, but which algorithm is more suitable for processing massive amounts of data is a constant concern. [Method] Select the isolation forest algorithm and the LOF algorithm. Firstly, analyze the principle and process of the two algorithms; then use the two algorithms to conduct experimental simulations through the data set to compare and study the accuracy and stability of the two algorithms in data anomaly detection. [Conclusion] The experimental data shows that the isolation forest algorithm is more suitable for anomaly detection in data mining; at the same time, improvements are proposed for the shortcomings of the isolation forest algorithm.