Πρόβλεψη διαλείψεων μεγάλης και μικρής κλίμακας στις ασύρματες επικοινωνίες με τεχνικές μηχανικής μάθησης

The objective of this thesis is the development of a robust system capable of predicting both large-scale and small-scale propagation losses in the context of a communication link between a low altitude UAV and a terrestrial terminal.The proposed system adopts a data-driven approach and is implemented with machine learning models, placing particular emphasis on the reliability and consistency of its predictions.Accurate and low-variance prediction of the wireless channel state enables dynamic allocation of communication resources, ensuring both optimal utilization and sustained communication connectivity across the link.To clearly illustrate the problem-solving workflow-starting from the initial problem formulation and data acquisition up to the final deployment of the prediction system-the standard machine learning system design pipeline is initially presented.Emphasis is placed on both the reasoning behind key design choices and on the sound theoretical justification of the adopted methodologies.Additionally, the thesis provides a detailed account of the real-world propagation measurements, aiming to produce a dataset that closely reflects the operational conditions the system is expected to encounter.The study's findings demonstrate that machine learning models can achieve highly accurate predictions of propagation losses, both in terms of average error and error variance.Particular attention is paid to the analysis of error dispersion, revealing that extreme prediction errors are notably constrained-a critical aspect, given that such extremes could potentially lead to link outages.Moreover, the fact that these results are obtained using real-world data suggests the practical viability of integrating such systems into future communication infrastructure.