The increasing need for machine learning to run with less computer resources is causing quantum computing to be considered as a viable substitute for conventional computing. This paper presents a hybrid quantum-classical system that improves learning by means of a classical optimisation loop combined with parameterised quantum circuits. While conventional procedures handle parameter updates and convergence, the proposed approach stores and processes high-dimensional data using quantum variational circuits. Qiskit is used by us to construct this architecture and run it on conventional datasets like Iris, MNIST (binary), CIFAR-100 (subset), HIGGS, and GTEx (gene expression). Particularly for jobs with complex, nonlinear patterns, the findings indicate that the mixed approach is as accurate as or more accurate than classical baselines. Real quantum hardware also works well with the system; noise has no impact on its precision. This work demonstrates how combined quantum-classical learning models might circumvent hardware constraints and provide genuine quantum benefits for applications requiring large data processing.