Turbulent convection flows are ubiquitous in natural systems such as in the atmosphere or in stellar interiors as well as in technological applications such as cooling or energy storage devices. Their physical complexity and vast number of degrees of freedom prevents often an access by direct numerical simulations that resolve all flow scales from the smallest to the largest plumes and vortices in the system and requires a simplified modelling of the flow itself and the resulting turbulent transport behaviour. The following article summarises some examples that aim at a reduction of the flow complexity and thus of the number of degrees of freedom of convective turbulence by machine learning approaches. We therefore apply unsupervised and supervised machine learning methods to direct numerical simulation data of a Rayleigh-Bénard convection flow which serves as a paradigm of the examples mentioned at the beginning.