QUANTUM-AWARE FEATURE SELECTION FOR QSVC AND QCBM CLASSIFIERS: ENCODING-EFFICIENT INFORMATION GAIN UNDER NISQ CONSTRAINTS

Srikumar Nayak

Authors

Srikumar Nayak Advanced Artificial Intelligence (AI) Engineering and Cybersecurity Threat Research, LTIMindtree Research, New York City, NY, USA.

Abstract

Variational quantum classifiers such as the Quantum Support Vector Classifier (QSVC) and the Quantum Circuit Born Machine (QCBM) inherit sharp qubit and depth budgets from noisy intermediate-scale quantum (NISQ) hardware, so the choice of which classical features to encode into a bounded register materially shapes both accuracy and runtime. Classical feature selectors such as principal component analysis and mutual information ignore the geometry of the quantum embedding and can waste scarce qubits on features the encoding cannot separate. This paper proposes Quantum-Aware Feature Selection (QAFS), a hybrid procedure that scores each candidate feature by a quantum Fisher information proxy computed on a shallow angle-encoded ansatz and selects a subset that maximizes label-conditional kernel dispersion under the target encoding. QAFS is evaluated on three public tabular classification benchmarks against random, PCA-based, and mutual-information selectors, using QSVC and QCBM as downstream classifiers under an eight-qubit budget. Across benchmarks the proposed framework raises validation accuracy from 0.812 (PCA + QSVC) to 0.883 (QAFS + QSVC) and F1-score from 0.741 to 0.842, while reducing mean kernel-evaluation cost by roughly 34 percent at n = 8 qubits. Robustness experiments show QAFS retains a clear margin under depolarizing noise up to p = 0.03, degrading more gracefully than encoding-agnostic baselines. The results indicate that feature selection tailored to the quantum encoding, not merely to classical variance, is the practically useful lever on NISQ devices, and that the sensitivity gain from Fisher-based scoring composes usefully with both kernel-based (QSVC) and generative (QCBM) heads.

Keywords

Quantum Machine Learning, Quantum-Aware Feature Selection, Quantum Support Vector Classifier (QSVC), Quantum Circuit Born Machine (QCBM), NISQ Encoding, Angle Encoding, Fisher Information, Variational Quantum Circuits.