*Result*: Coarse-to-Fine Positioning Method Using Graph Neural Networks with Feature Reduction Analysis for Indoor Positioning Systems.

*The emergence of Location-Based Services has resulted in a demand for a robust and applicable positioning system. While fingerprinting methods are generally more robust compared to geometric methods, the inability to adapt with changes in the environment became a major applicability weakness for the fingerprinting methods. In this research, we proposed a novel fingerprint-based positioning system with Weighted K-Nearest Neighbors for coarse positioning to build a graph containing spatial information of the fingerprint map. We used the fingerprint graph to train a Graph Neural Networks model for fine positioning. The results showed that the proposed model performed better compared to the baseline models, achieving the best accuracy of 55.39cm Mean Positioning Error for the proposed model using GraphSAGE layer. We also implemented transmitter removal as a feature reduction method to simulate the proposed method and baseline models performance in environments with low number of transmitters. The findings showed that the feature reduction caused the Mean Positioning Error of the models to increase, with each model having different rate of increase. The overall results from the feature reduction process showed that the proposed method maintained better Mean Positioning Error values but had greater rate of change in Mean Positioning Error values. [ABSTRACT FROM AUTHOR]*