A quantum feature selection framework via ground state preparation

Abstract

Abstract Traditional feature selection methods face the challenges of increasing time complexity and local optima. In previous works, many classical feature selection methods were accelerated through quantum algorithms. However, these approaches still inherit the constraints of these classical methods as they do not address the issue of local minima. Here, we propose a novel quantum feature selection framework based on the classifier’s result, which utilizes Hamiltonian encoding and a ground state preparation algorithm. Numerical experiments are conducted on real-world datasets from the finance and medicine domains. Moreover, the results demonstrate that the proposed method produces the same or better classification accuracy on the classifier than the original data without feature selection. Overall, our approach presents a promising solution to feature selection using quantum computing.