Berke KilincBerkay HabibVolkan Rodoplu2025-10-062023979-8-3503-1161-7, 979-8-3503-1162-42769-400310.1109/WF-IOT58464.2023.10539438http://dx.doi.org/10.1109/WF-IOT58464.2023.10539438https://gcris.yasar.edu.tr/handle/123456789/6183In this work we propose a novel machine learning-based architecture that processes the Channel Impulse Response (CIR) for mobile Internet of Things (IoT) indoor localization. Our architecture is comprised of three stages: First it pre-processes the Channel Impulse Response of the channel from the mobile device to each anchor by lumping the channel tap values at a configurable resolution. Second the Machine Learning-Based Multipath Profile Processing block applies feature selection to the pre-processed channel taps. Third in the Machine Learning Based Feature Fusion block the selected features are combined to estimate the position of the mobile device. In order to test the performance of our architecture we use two distinct datasets that were collected in home and office environments respectively. The results demonstrate that our work can significantly improve indoor localization accuracy. This work paves the way to significant performance improvements in indoor localization by processing the Channel Impulse Response via machine learning algorithms.EnglishIndoor Localization, Machine Learning (ML), Internet of Things (IoT), Channel Impulse Response(CIR), Ultra-Wideband(UWB), Multipath Profile, Indoor PositioningConference Object