Performance prediction of visible light communication system using machine learning: a comparative analysis of classifiers

Abstract The performance of a VLC system depends heavily on various parameters, such as link distance, transmitter and irradiance half-angles, detection surface area, optical concentration factor, and incidence angle. However, predicting the system’s performance in real-world conditions based on combinations of these parameters is challenging. To address this, this paper proposes a machine learning-based prediction model that assesses the feasibility of VLC system configurations using known parameter values. A synthetic dataset was generated through the optical simulation software Optisystem, and multiple machine learning algorithms were trained on this data. The classifiers, including multilayer perceptron (MLP), random forest classifier (RFC), logistic regression (LR), naïve Bayes (NB), decision tree (DTR), and K-nearest neighbors (K-NN), were compared for this task. Results show that both MLP and RFC (with 50 estimators) achieved the highest accuracy of 95 %. However, MLP is preferred due to its superior cross-validation score of 84 and an ROC AUC of 98, making it the most effective model for predicting VLC system performance.