Quantum Machine Learning (QML) promises to combine quantum resources with machine learning to build more expressive or efficient models, but current noisy intermediate-scale quantum (NISQ) devices impose severe resource and noise constraints. This work delivers a structured, application-oriented survey complemented by a concise taxonomy of the major QML families. We further synthesize representative studies into a comprehensive table that systematically maps algorithms to their corresponding datasets, evaluation metrics, software frameworks, and hardware backends. Our comparative analysis reveals that while Quantum Support Vector Machines (QSVMs) excel at tabular and healthcare tasks and Quantum Neural Networks (QNNs)/Variational Quantum Circuits (VQCs) span diverse domains, both approaches face significant constraints: trainability bottlenecks, hardware noise, and reliance on small-scale simulator-based evaluations. We therefore advocate for unified benchmarking suites, hybrid classical–quantum workflows, noise-aware ansatz optimization, and realistic hardware-based protocols to accurately assess the practical viability of near-term QML systems.