Many efforts have been dedicated to establish, by fitting a large number of experimental data, zero-dimensional scaling laws for conventional scenarios to give global information on transport. Despite its advantages in designing future devices, this approach is not suitable to investigate transport phenomena in advanced scenarios. The present work focuses on the local thermal energy transport in JET discharges with and without ITBs. The aim is twofold: (i) to exhibit the dependencies of the heat diffusivity on local dimensionless parameters which seem to be relevant from a theoretical basis and (ii) to propose a purely empirical approach for predictive transport modelling. Neural networks will be introduced as an appropriate tool to handle such a problem and provide an alternative approach to previous attempts of approximating semi-empirically the experimental heat diffusivity function.