Comparison of Multi-Label Classification and Regression Models for Port Selection in Fluid Antenna Systems

Fluid antenna systems offer significant potential to enhance spatial diversity, which can substantially improve the quality of received signals. However, achieving such signal quality depends on selecting the port that maximizes the signal-to-noise ratio out of a large number of ports, which is a time-and resource-consuming process. Therefore, selecting the optimal port efficiently is crucial for maintaining high-performance communication. Consequently, this work presents a comparative study of machine learning approaches, particularly multi-label classification (MLC) and regression, for the first time to predict the best port based on limited observations. Results show that deep learning MLC and regression models optimized using the Optuna hyperparameter optimization framework can effectively predict the optimal port out of a few observations, with MLC ones slightly outperforming regression models in terms of the outage probability.