A Differentially Private Stacked Denoising Autoencoder-Based Fingerprint Localization Model in a Cloud Environment

Indoor localization services in cloud environments face two key challenges: protecting user location privacy during crowdsourced Wi-Fi fingerprint sharing and maintaining localization accuracy when learning from high-dimensional and noise-perturbed received signal strength (RSS) data. To address these challenges, this paper proposes DP-SLocCE, a differentially private fingerprint-based indoor localization framework built on stacked denoising autoencoders. The proposed framework adopts shuffled differential privacy to provide user-level privacy protection during data transmission, while leveraging stacked denoising autoencoders to perform robust feature extraction and progressive dimensionality reduction from noisy RSS fingerprints. Extensive experiments on three real-world indoor localization datasets demonstrate that, under differential privacy constraints, DP-SLocCE consistently outperforms mainstream deep learning models in localization accuracy. Moreover, in high-dimensional scenarios, it achieves accuracy comparable to CNN-based approaches while significantly reducing training and testing time, highlighting its practicality for cloud-based indoor localization services.