WiFi Indoor Localization Based on XGBoost with Access Point Reliability Weighting in Multi-Device Environments
WiFi Indoor Localization Based on XGBoost with Access Point Reliability Weighting in Multi-Device Environments
Mahsa Maleki1 Maryam Chinipardaz2
1) B.Sc. Graduate in Computer Engineering, Jundi-Shapur University of Technology Dezful, Dezful, Iran -
2) Assistant Professor, Faculty of Electrical and Computer Engineering, Jundi-Shapur University of Technology, Dezful, Iran -
IOI Code
:
AAHD-ABKDAZ
Publication :
National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
WiFi fingerprint-based indoor localization remains challenging in real-world, multi-device settings due to temporal signal fluctuations, uneven access point (AP) coverage, and hardware heterogeneity across devices. Most prior studies have focused primarily on reducing average error, giving less attention to model stability under critical errors, device-specific behavior, and the role of AP reliability. In this study, the performance of four algorithms—WKNN, Random Forest, MLP, and XGBoost—was evaluated on the real, multi-device TUJI1 dataset, and in addition to conventional accuracy metrics, stability indicators including the 95th percentile error (P95) and the cumulative distribution function (CDF) of error were analyzed. An AP Reliability Weighting (AP-RW) method was then proposed, which determines the weight of each AP based on its detection rate and signal stability, and is incorporated into the XGBoost model without discarding any AP. Results showed that XGBoost achieved the best average accuracy among the baseline models, with a mean error of 2.0081 m, while the XGBoost+RW variant maintained a comparable mean error while reducing the 95th percentile error from 4.5424 m to 4.3474 m. Device-wise analysis further revealed that hardware heterogeneity remains an open challenge. These findings indicate that AP reliability weighting is more effective at improving stability and reducing critical errors than at reducing average error.
Keywords : indoor localization wifi fingerprinting xgboost ap reliability weighting tuji1