Article
Privacy-Enhanced Key-Value Data Collection for Large-Scale IoT Environments
Privacy-preserving data collection is essential for IoT environments, where massive volumes of key-value data are continuously generated and transmitted across distributed devices. Local Differential Privacy (LDP) provides protection without requiring trusted data collectors, but conventional mechanisms often introduce substantial estimation errors and reduced data utility, particularly for composite key-value structures. This work addresses these limitations through an accuracy-oriented privacy-preserving collection framework using a key-value dataset containing categorical keys and associated values collected over a defined observation period. The data is preprocessed by removing missing and duplicate records, validating key-value pairs, encoding categorical attributes, and applying normalization where required. The implemented models include Randomized Response, Generalized Randomized Response, Optimized Unary Encoding, and proposed hybrid mechanisms combining key and value perturbation with sampling-based processing. Performance is evaluated using privacy level, hit rate, estimation variance, precision, and utility preservation. Among the evaluated approaches, the proposed hybrid key-value perturbation mechanism achieves the highest hit rate and the lowest estimation variance, demonstrating improved estimation reliability under privacy constraints. The results indicate that coordinated sampling and perturbation can provide stronger privacy protection while maintaining high data utility, offering an effective approach for accurate and privacy-preserving keyvalue data collection in IoT environments.
Full Text Attachment
