Analyzing the Performance of Radiolocalization Algorithms using Data Augmentation

This work aims to investigate and compare the impact of different synthetic data generation techniques on the performance of fingerprint-based localization models. We conducted experiments on two databases, exploring conditional, non-conditional, selective, and non-selective synthetic data generation methods. Three machine learning-based localization models were utilized, resulting in a total of 90 models being trained-some using only real data while others incorporated synthetic data. The results indicate that synthetic data generation can enhance the performance of machine learning prediction models, particularly for those based on support vector regression. Additionally, the conditional and selective generation methods outperformed their non-conditional and non-selective counterparts.