Problem
Research questions and friction points this paper is trying to address.
Classification of high-dimensional low sample size data
Development of robust, parameter-free classifiers
Performance comparison with existing methods
Robust Classification of High-Dimensional Data using Data-Adaptive Energy Distance
🤖 AI Summary
To address the sensitivity of high-dimensional, low-sample-size (HDLSS) classification to noise, outliers, and distributional shift, this paper proposes a parameter-free, distribution-agnostic robust classifier. The method innovatively couples energy distance with local geometric structure for adaptive distance scaling; integrates data-driven kernel bandwidth selection, weighted energy distance computation, subspace sparsity-aware projection, and a distance-based nearest-neighbor framework. Crucially, it imposes no distributional assumptions—neither Gaussianity nor moment conditions—and significantly enhances discriminative power under small-sample and non-spherical cluster settings. Evaluated on 12 HDLSS benchmark datasets, it achieves an average accuracy improvement of 3.2% over state-of-the-art methods. It maintains ≥89% accuracy under 50% label noise and operates two orders of magnitude faster than deep learning baselines during inference.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Data-Adaptive Energy Distance
Parameter-free Classifiers
HDLSS Data Robustness
🔎 Similar Papers
No similar papers found.
