5G Technology: ML Hyperparameter Tuning Analysis for Subcarrier Spacing Prediction Model

Abstract

Resource optimisation is critical because 5G is intended to be a major enabler and a leading infrastructure provider in the information and communication technology sector by supporting a wide range of upcoming services with varying requirements. Therefore, system improvisation techniques, such as machine learning (ML) and deep learning, must be applied to make the model customisable. Moreover, improvisation allows the prediction system to generate the most accurate outcomes and valuable insights from data whilst enabling effective decisions. In this study, we first provide a literature study on the applications of ML and a summary of the hyperparameters influencing the prediction capabilities of the ML models for the communication system. We demonstrate the behaviour of four ML models: k nearest neighbour, classification and regression trees, random forest and support vector machine. Then, we observe and elaborate on the suitable hyperparameter values for each model based on the accuracy in prediction performance. Based on our observation, the optimal hyperparameter setting for ML models is essential because it directly impacts the model’s performance. Therefore, understanding how the ML models are expected to respond to the system utilised is critical.

Keywords

• machine learning
• hyperparameter tuning
• 5G
• resource allocation
• resource management