講演情報
[B-1C-37]Effects of Decision Threshold and Neighbor-Size Parameters on Adaptive Hybrid Localization Performance
〇Koularp THONGSAVANH1, Minseok KIM1 (1. Niigata University)
キーワード：
Indoor localization、Angle-of-arrival (AoA)、CSI fingerprinting、Sensitivity analysis、Adaptive method selection
In our previous work, we showed that AoA measurements are subject to outliers and evaluated the outlier robustness of AoA-based localization, including outlier sparsity-promoting linear regression (OSPLR) [1]. Although OSPLR is designed for outlier robustness, it can still degrade under severe or simultaneous multi-sensor outliers [1]. This motivated a reliability-based adaptive framework that falls back to CSI fingerprint-based weighted K-nearest neighbor (WKNN) localization [2] whenever AoA is unreliable. However, this framework depends on three parameters, namely the threshold scale β, classifier neighbor count K, and WKNN neighbor count KW, which were chosen empirically without a clear understanding of their sensitivity. This paper addresses that gap by systematically evaluating the sensitivity of all three parameters, clarifying which require careful tuning and which can be chosen flexibly.