A machine learning enabled link adaption (LA) and scheduling framework is presented for Industrial Internet of Things (IIoT), leveraging quasi-periodicity of traffic in IIoT. The following steps are introduced: i) a reduced complexity link establishment accounting jointly for beamforming and load management; ii) interference prediction using long short-term memory neural networks; iii) semi-coordinated scheduling based on node grouping for interference avoidance. Through numerical evaluation it is demonstrated that the proposed approach can substantially improve average spectral efficiency by as much as 62% in a realistic IIoT scenario at negligible overhead.