Data Re-uploading in Quantum Kernels: A Hybrid Approach for Tabular Data Classification

Quantum Machine Learning (QML) has emerged as a promising paradigm for enhancing classical learning models by leveraging the expressive power of quantum Hilbert spaces. In this work, we propose a novel quantum kernel design for Support Vector Classification (SVC) based on data re-uploading—a layered circuit encoding strategy that injects classical features multiple times across quantum layers to enrich model expressivity without increasing the number of qubits. To ensure compatibility with current Noisy Intermediate-Scale Quantum (NISQ) devices, we apply Principal Component Analysis (PCA) to reduce input dimensionality prior to quantum embedding. The resulting quantum kernel is computed via fidelity between quantum states and integrated into a hybrid quantum-classical SVC framework. We evaluate the proposed method on three tabular datasets: Iris, Wine, and Breast Cancer Wisconsin (Diagnostic), and benchmark it against classical SVCs (linear, RBF) and static quantum kernels. Our results demonstrate consistent improvements, with up to 3.6% higher accuracy and 2.3% gain in F1-score. Ablation studies highlight the trade-off between re-uploading depth and performance, and t-SNE visualizations reveal improved class separability in the quantum feature space. These findings show that data re-uploading offers a practical and scalable enhancement to quantum kernel methods, advancing their applicability under realistic hardware constraints.