In this article, machine learning is used to predict lifetime under isothermal low-cycle fatigue and thermo-mechanical fatigue loading, both of which represent the most complex loadings that couple creep, fatigue and oxidation damage. A uniaxial fatigue and fatigue–creep dataset, which was obtained for temperatures of between 300°C and 600°C for a low-alloy martensitic steel, is utilized in this study. Two different machine learning based approaches to lifetime prediction are demonstrated. The first approach is based only on a shallow neural network, whereas the second approach is proposed as a combination of a sequence learning based model – either long short-term memory network or gated recurrent unit – with the shallow neural network. A good correlation between the experiment and the prediction suggests that lifetime under complex thermo-mechanical loading can be reasonably predicted via the proposed machine learning based damage models.