Deep Learning-Enhanced Single Station Wi-Fi Localization for Social Networking in Smart Environments

This paper addresses the challenge of indoor localization using a single Wi-Fi access point (AP), proposing a cost-effective alternative to traditional multi-AP systems. Our objective is to demonstrate that single-AP localization can achieve high accuracy through advanced data processing and deep learning techniques, significantly reducing hardware requirements and deployment complexity. We introduce a comprehensive methodology that enhances localization precision through standardized feature scaling, addressing fundamental challenges including signal noise, environmental variability, and multipath interference. The research evaluates four neural network architectures incorporating class weighting, SMOTE oversampling, focal loss functions, and balanced batch sampling to mitigate class imbalance issues. Our experimental results demonstrate that a deep neural network combining SMOTE with focal loss achieves superior performance, with macro-F1 scores exceeding 0.97 and accuracy above 95% in controlled environments. We provide detailed computational complexity analysis, showing that our approach maintains efficiency despite its sophistication, with inference times under 5ms on standard hardware. The paper also addresses practical deployment considerations and limitations, offering strategies for adapting to changing environments. Our findings confirm that single-AP Wi-Fi localization, when enhanced by robust data curation and specialized deep learning, offers a viable, flexible, and cost-efficient solution for indoor positioning across various smart environment applications.