Quantum Support Vector Machine Algorithm

Abstract

Churn prediction of customers is highly significant in banking as the cost of retention is much lower than the cost of acquisition. Usually individuals operate with the standard ML frameworks, such as SVM, although they have a ceiling when the data is of high dimension and extremely non-linear - they become more and more sluggish and are inadequate to actually model all the complexity. The paper presents a Quantum Support Vector Machine (QSVM) that is used to predict the customers who are going to leave the bank. Unlike The preprocessing in classical code, such as clean up the data, encoding of categories, scaling and then PCA to reduce the feature set to one that fits within the small number of qubits, all proceeds as normal. Then we apply the shrunk data to a quantum state, using ZZFeatureMap, and compute a quantum kernel which informs us of the similarity between data points in Hilbert space. Initialization of the QSVM in Python using Qiskit and then running it on the Qiskit Aer simulator. It is reasonably good: overall accuracy of around 88 percent and good scores on non-churn customer identification and fair adequacy in identifying churners. These figures demonstrate that quantum-enhanced ML might be a viable alternative to quantum hardware in the realm of real business analytics and is going to become even more feasible as quantum technology advances.