Quantum Algorithm for Unsupervised Anomaly Detection

Abstract

Anomaly detection, an important branch of machine learning, plays a critical role in fraud detection, health care, intrusion detection, military surveillance, etc. As one of the most commonly used unsupervised anomaly detection algorithms, the Local Outlier Factor algorithm (LOF algorithm) has been extensively studied. This algorithm contains three steps, i.e., determining the k-distance neighborhood for each data point x, computing the local reachability density of x, and calculating the local outlier factor of x to judge whether x is abnormal. The LOF algorithm is computationally expensive when processing big data sets. Here we present a quantum LOF algorithm consisting of three parts corresponding to the classical algorithm. Specifically, the k-distance neighborhood of x is determined by amplitude estimation and minimum search; the local reachability density of each data point is calculated in parallel based on the quantum multiply-adder; the local outlier factor of each data point is obtained in parallel using amplitude estimation. It is shown that our quantum algorithm achieves exponential speedup on the dimension of the data points and polynomial speedup on the number of data points compared to its classical counterpart. This work demonstrates the advantage of quantum computing in unsupervised anomaly detection.