As the parameterisation of an electrochemical model is time-consuming and requires a lot of equipment, this paper presents a new approach for determining the 18 main parameters of a Doyle-Fuller-Newman model. This is based on a machine learning model trained with simulations. By using PyBaMM 20000 EIS are simulated and used as a training dataset for the feed-forward machine learning model. The model gets as an input the imaginary and real parts of the EIS and has as the output the values of the parameter. The cell used in the state-of-the-art (SoA) parameter set (LGM50) is compared with the almost identical successor cell (LGM50T). For this work, the most easily determined physical parameters were determined in a post-mortem analysis. Comparing simulated and measured EIS, the EIS simulation using the estimated parameter set fits the measured data better than the SoA one, especially for higher frequencies. However, the diffusion is not satisfactorily modelled in the estimated parameter set, but still better than in the SoA. The approach presented here is faster, less complex, easily adaptable and results in a good model behaviour compared to state-of-the-art parameterisation methods from the literature. However, in the current state, it is not possible to determine the physical parameters. To further improve the results, GITT measurements should be used to determine the diffusivity.