Over parameterisation in classical machine learning is a phenomenon that has shown benefits in training and generalisation abilities. There exists an adjacent theory for quantum circuits, that provides well-defined metrics for measuring and calculating the onset of over-parameterisation. However, this framework has mostly been applied to Hamiltonian simulation problems, where the number of trainable parameters to solve the problem is well known. In this presentation, we present results on learning a subset of MNIST-1D, where the architecture of the quantum circuit allows for over-parameterisation. In addition, we present results in learning and generalisation capabilities, and what quantum properties the circuit utilises.