Millimeter Wave (mmWave) technology coupled with full duplex (FD) communication has the potential of increasing the spectral efficiency. However, the self-interference (SI) encountered in the FD mode and the ubiquitous multi-user interference (MI) contaminates the signal. Furthermore, the system performance may also be limited by channel aging that arises because of the time-varying nature of the channel. Therefore, in this paper, we conceive FD hybrid beamforming (HBF) for K-user multiple-input multiple-output (MIMO)-aided orthogonal frequency division multiplexing (OFDM) using learning-aided channel prediction. We first derive a joint precoder and combiner design for full duplex K-user MIMO-OFDM interference channels, where we aim for minimizing both the residual SI and the MI, followed by an iterative hybrid decomposition technique developed for OFDM systems. Then, we propose a learning-aided channel prediction technique for systems suffering from channel aging relying on a radial basis neural network, where we show by simulation that upon using sufficient training, learning-assisted channel prediction can faithfully estimate the current channel. Furthermore, we demonstrate by simulations that our proposed joint hybrid precoder and combiner design outperforms the popular Eigen beamforming (EBF) technique by about 5 dB for a 128 × 32-element MIMO aided OFDM system having 32 sub-carriers.