Title: Comparative Study of Classical and Quantum Learning Models: Performance and Embedding Techniques in Binary Classification

This study compares the performance of classical and quantum computational strategies in binary classification tasks. We evaluate traditional gradient-based models, including neural networks and logistic regression, against the nascent Variational Quantum Circuits. Our evaluation uses concrete metrics, such as the Fisher Information Metric, accuracy, and the F1 score. To standardize the comparison, we assign an equivalent number of parameters, utilize the effective dimension capacity measure, and employ a synthetic dataset tailored for binary classification to avoid biases. We also investigate the potential of various quantum embedding methods such as angular, amplitude, stereographic, etc. Our study provides additional evidence supporting the benefits of quantum-enhanced learning methods to model capacity. Our results pave the way for a deeper understanding of applying learning theory to emerging computational paradigms.