Quantum Feature Pruning for Scalable and Efficient Quantum Kernel-Based High-Dimensional Classification

Quantum Kernel Methods (QKMs) have shown promising potential in enhancing classification tasks through quantum-enabled feature mapping. However, their practical adoption is hindered by high computational complexity, poor scalability in high-dimensional spaces, and sensitivity to quantum noise. This paper introduces a novel framework, Quantum Feature Pruning (QFP), designed to optimize QKMs by selectively retaining only the most informative quantum-transformed features. The primary objective of this work is to enhance the computational efficiency, scalability, and noise robustness of QKMs without compromising classification accuracy. The proposed QKM-QFP model is evaluated on three benchmark datasets-Ionosphere, Fashion-MNIST, and a Synthetic HighDimensional Dataset-and compared against classical and quantum kernel baselines. Experimental results demonstrate that QKM-QFP achieves up to 92.5% accuracy, reduces computation costs by up to 31%, and significantly improves noise resilience and high-dimensional performance. This study highlights the practicality of QFP-enhanced QKMs for real-world quantum machine learning and paves the way for future work on adaptive pruning and hardware-aware optimization.