Empirical Performance Data for Quantum vs Classical SVM on MNIST, Fashion-MNIST, and Breast Cancer Datasets

This dataset contains the experimental results and visual benchmarks for a systematic analysis of Empirical Quantum Advantage in supervised machine learning. The research compares the performance of a Quantum Support Vector Classifier (QSVC) against a classical Radial Basis Function (RBF) SVC baseline across three distinct datasets: MNIST, Fashion-MNIST, and the Breast Cancer Wisconsin (Diagnostic) dataset. Experimental Objectives: The study investigates how quantum feature mapping (specifically the ZZFeatureMap) influences classification accuracy, F1-score, and computational overhead as a function of training sample size (N). By keeping the feature dimensionality constant (via PCA reduction to 5 components), the experiment isolates the impact of the quantum kernel's inductive bias. Data Content: Datasets: The raw atasets used in the experiment Numerical Results: Comma-Separated Values (.csv) files containing mean performance metrics, standard deviations, and statistical significance indicators (p-values) derived from the Nadeau-Bengio Corrected T Tests. Visual Summaries: Multi-panel PDF plots illustrating learning curves and training time ratios for each dataset. Metadata: Comprehensive column definitions and hyperparameter configurations are provided in the accompanying README.md file. Technical Context: The quantum components were implemented using IBM Qiskit, utilising the FidelityQuantumKernel and simulated on the qiskit_aer backend. Classical baselines were implemented using Scikit-learn. This data is intended to support research into the practical feasibility of NISQ-era quantum machine learning algorithms on structured and unstructured datasets.