A novel privacy-preserving distributed anomaly detection method

Abstract

Anomaly detection refers to the algorithm to find the anomalies among the data. As a branch of data mining, it has important research significance. With the advance of sensor technology, data is always distributed at many places. To ensure that the data owners privacy data is not disclosed in the process of anomaly detection, the privacy preserving scheme is necessary. In this paper, we propose a provable secure structure, Secure Isolation Forest(SIF), which is a distributed anomaly detection algorithm based on ensemble isolation principle. We improve performance and detection capabilities by fixed the height of trees and adopt an effective homomorphic cryptosystem. Our construction allows the inputs encrypted by different independent public keys. Lastly, we highlight the practicability of our construction by extensive experimental evaluation.