This study analyses the performance of the stacked long short-term memory (S-LSTM)-based non- orthogonal multiple access (NOMA) system under independent and non-identically distributed (i.n.i.d.) Nakagami-m fading channel links. The NOMA system has been used in conjunction with the multiple-input multiple- output (MIMO) scheme to achieve the diversity gain. The proposed deep learning (DL) receiver employs the singular value decomposition (SVD) scheme to get the optimal performance for the MIMO-NOMA system. In addition, the effectiveness of the proposed system is evaluated by analysing the effect of various shape parameter values, sample sizes, learning rates, and pilot symbols (PS). In addition, the performance of the proposed receiver is compared to that of conventional MIMO-NOMA receivers. By simulating various channel conditions, it is demonstrated that the performance loss due to the i.n.i.d. fading connection assumption is small under the worst fading channel circumstances and increases as the channel condition improves. The simulation results and the analytical results are in close agreement.