Key–value data collection with local differential privacy for urban air quality monitoring in crowdsensing

The growth of IoT and mobile devices has led to Mobile Crowdsensing (MCS), a cost-effective data collection method crucial for smart cities. While MCS outperforms wireless sensor networks, it may expose workers’ sensitive data, such as location and identity, in air quality monitoring. Traditional privacy-preserving techniques, such as location obfuscation and data perturbation, have inherent limitations in ensuring strong privacy protection. Moreover, the frequent uploading of numerical data during task execution requires a larger privacy budget, thereby increasing the risk of privacy leakage. To solve these problems, this paper proposes a key–value data collection scheme based on local differential privacy for air quality monitoring in smart cities. The proposed scheme aims to protect user privacy while ensuring data utility. It consists of two main phases: data collection and data prediction. During the data collection phase, workers locally perturb both the task location (key) and the sensed data (value), utilizing the correlation between keys and values to enhance data utility. The system subsequently aggregates the perturbed data and applies bias correction to ensure unbiased estimation. In the prediction phase, an exponential smoothing technique is introduced to mitigate the impact of privacy-preserving mechanisms on prediction accuracy. This method effectively reduces random fluctuations in the data, thereby enhancing the overall prediction performance. Experiments on real-world datasets show that the proposed scheme outperforms other privacy-preserving algorithms in efficiency while maintaining nearly the same prediction accuracy as non-privacy-preserving methods, effectively balancing privacy and data utility.