NOMA Performance Prediction and Impact Analysis of System Parameters through a Supervised Machine Learning Approach

This paper presents a supervised machine learning approach designed to predict the performance of Non-Orthogonal Multiple Access (NOMA) and Orthogonal Multiple Access (OMA) systems, focusing on key system parameters such as user density, total bandwidth, maximum transmit power, and subband aggregation. Our objectives are twofold: firstly, to a priori determine the most suitable access method - NOMA or OMA - under varying network conditions, and secondly, to ascertain the impact of each system parameter on performance. The unique contribution of our work lies in its capacity to forecast the performance outcomes of NOMA and OMA prior to allocation, thereby informing the selection of the most appropriate access scheme. Our findings highlight the critical influence of user density on system performance, revealing that a high number of users per cell can reduce NOMA's effectiveness in comparison to OMA. By employing this dual-faceted approach, we enable precise system optimizations, ensuring the best possible performance under diverse operational scenarios.