A central problem of development in chemical and pharmaceutical industries is modelling a cheap to evaluate surrogate function, that approximates a given black box function sufficiently well. As state-of-the-art methods from classical machine learning struggle to solve this problem accurately for the typically scarce and noisy datasets in practical applications, investigating novel approaches is of great interest to chemical companies worldwide. We demonstrate that quantum neural networks (QNNs) offer a particularly promising approach to this issue and experimentally support recent theoretical findings indicating their potential to outperform classical equivalents in training on small datasets and noisy data. Our contribution displays the first application centered exploration of using QNNs as surrogate models on higher dimensional, realistic data. In extensive experiments, our QNN significantly outperforms a minimalist classical artificial neural network on noisy and scarce data, displaying a possible advantage of quantum surrogate models empirically. Finally, we demonstrate the performance of current NISQ hardware experimentally and estimate the gate fidelities necessary to replicate our simulation results.