Machine learning-based methods for MCS prediction in 5G networks.

In the ever-evolving landscape of wireless communication systems, including fifth-generation (5G) networks and beyond (B5G), accurate Modulation and Coding Scheme (MCS) prediction is crucial for optimizing data transmission efficiency and quality of service. Traditional MCS selection methods rely on predefined rules and heuristics, offering transparency and control but lacking adaptability in changing wireless conditions. The emergence of Machine Learning (ML) has brought transformative capabilities, particularly in MCS prediction. ML leverages data-driven models, promising improved accuracy and adaptability in dynamic wireless environments. This paper marks a novel endeavor in this domain, as it explores and evaluates a range of machine learning (ML) techniques for predicting MCS in orthogonal frequency-division multiplexing (OFDM) systems, representing the first such investigation in this field. Additionally, it introduces a specialized Deep Neural Network (DNN) architecture with two hidden layers for MCS prediction, guided by performance metrics such as accuracy, precision, recall, and F1-score. The examined ML methods include Artificial Neural Networks (ANN), Support Vector Machine (SVM), Random Forest (RF), and Bagging with k-NN (B-kNN). These methods undergo thorough training and evaluation using a dataset generated from simulations of non-standalone 5G networks. The study incorporates physical layer measurements and employs a ray-tracing path loss prediction model for comprehensive environmental characterization. Also, advanced data mining techniques preprocess raw data, addressing model underfitting and overfitting challenges. Finally, performance evaluation results reveal that the ANN with two hidden layers achieves the highest accuracy at 98.71%, while RF and B-kNN methods attain the lowest accuracy, below 88.65%. SVM and ANN models, with one and four hidden layers, respectively, demonstrate comparable MCS prediction accuracy, ranging from 97.02 to 97.30%. [ABSTRACT FROM AUTHOR]