Machine Learning Techniques Benchmarking based on Global optimization Methods for Parameter Estimation of Log-distance Path loss model

Abstract

 Accurate parametric estimation of path loss models using distinctive machine learning techniques are crucial for ensuring reliable mobile communications. They inform network planning, help optimize resource allocation, and ultimately enhance user experience. Traditional methods of parameter estimation often struggle to adapt to the complex, dynamic environments encountered in real-world scenarios, leading to inaccuracies that can compromise network efficacy. To address these challenges, machine learning-based global optimization methods have emerged as a promising alternative, offering sophisticated techniques that can enhance the precision of parameter estimation. This paper explores the benchmarking of various machine learning algorithms based global optimization method in predictive path loss modelling at three different study locations. The methods include the Particle Swarm Optimization (PTS), Pattern search (PATS), Genetic Algorithm (GA), and Simulated Annealing (SIA). With mean square error evaluation metric, the results reveals that PTS attains better global precision credibility with lower MSE values as the iteration number increases when estimating the log-distance model parameters. The results attained by PTS clearly showcase its efficiency in finding optimal or near-optimal solutions in continuous search domains. Its adaptive nature allows it to quickly converge on good solutions, making it more suitable for the parametric path loss model estimations.