Predicting Water Quality using Quantum Machine Learning: The Case of the Umgeni Catchment (U20A) Study Region

Abstract

In this study, we consider a real-world application of QML techniques to study water quality in the U20A region in Durban, South Africa. Specifically, we applied the quantum support vector classifier (QSVC) and quantum neural network (QNN), and we showed that the QSVC is easier to implement and yields a higher accuracy. The QSVC models were applied for three kernels: Linear, polynomial, and radial basis function (RBF), and it was shown that the polynomial and RBF kernels had exactly the same performance. The QNN model was applied using different optimizers, learning rates, noise on the circuit components, and weight initializations were considered, but the QNN persistently ran into the dead neuron problem. Thus, the QNN was compared only by accraucy and loss, and it was shown that with the Adam optimizer, the model has the best performance, however, still less than the QSVC.