Leveraging machine learning for space division multiplexing performance: a classifier-focused evaluation

Abstract This paper explores space division multiplexing as a next-generation solution for optical fiber communication systems. The purpose of this paper is to investigate how well machine learning algorithms predict the performance of SDM systems. Using simulations of SDM systems, the study evaluates classifiers such as multilayer perceptron, random forest classifier, logistic regression, naive Bayes, decision tree, and k-nearest neighbors. The performance of the classifiers is measured in terms of accuracy, cross-validation scores, and ROC (area under the curve) (AUC). The findings from dataset 1 show that MLP and KNN reach the maximum accuracy of 93. Though logistic regression has the best cross-validation score of 87, MLP is still chosen due to its superior ROC score of 99. In dataset 2, MLP achieves the maximum accuracy of 84. MLP is favoured because of its higher cross-validation score of 79 obtained using logistic regression, compared to its superior ROC score of 92. Although logistic regression exhibits good cross-validation performance, the study prefers MLP because of its superior accuracy and remarkable ROC-AUC scores in both datasets, indicating its robust classification abilities across a wide range of data distributions.