A machine-learning-based labelling diversity model for predictive analysis: using 16QAM as a case study

Abstract

The recent advancement and enhancement that optimized uncoded space-time labelling diversity (USTLD) have provided significant diversity gains. By adopting the use of evolutionary algorithms, labelling diversity (LD) mapper designs produced are near-optimal in quality. The only disadvantage to the use of evolutionary algorithms is that the produced solution is not always optimal. To ease the calculation of how much a mapper design had achieved LD, this paper proposes a machine learning-based analysis to predict the amount of LD achieved by a mapper. In this paper, only the 16QAM constellation is studied as a simple case. Six machine learning-based algorithms were proposed in this paper, namely multi-linear regression (MLR), support vector regression (SVR), decision trees (DT), random forest (RF), K-nearest neighbours (KNN) and a simple artificial neural network (ANN). From the results obtained from the experiments, it can be seen that the MLR algorithm is the least time complex while the ANN is the most time complex. It is also important to note that the DT and KNN algorithms take a comparatively short amount of time to execute. When compared in terms of machine learning metrics, it was shown that the ANN algorithm performed the best with the least amount of error while the MLR algorithm performed the worst with the highest amount of error. Thus, it could be seen that the results from this paper provide a positive outlook on applying machine learning algorithms to the LD problem.