International Joint Conference on Neural Networks (IJCNN)
Semi-Supervised Multichannel Anomaly Detection Using QAOA-Driven Selection and Set Covering
Marco Casalbore, Leonardo Lavagna, Antonello Rosato, Francesca Cuomo, Massimo Panella
Abstract
Anomaly detection in multivariate time series remains a challenging task due to high dimensionality, temporal dependencies, and complex inter-variable relationships. In this work, we extend a previously proposed hybrid quantum-classical anomaly detection framework to the multivariate setting. The method formulates anomaly detection as a Quadratic Unconstrained Binary Optimization problem solved via the Quantum Approximate Optimization Algorithm, where multichannel information is implicitly encoded in a problem-dependent cost matrix. The proposed method is validated on a real-world dataset of water smart meter measurements, demonstrating robust and competitive detection performance. The goal is to also enable empirical analysis of solution multiplicity and near-degeneracy effects in the quantum optimization landscape, along with their impact on the induced geometric representations of normality.