Adaptive Methods for Improving the Accuracy of Truth Discovery Considering Data Characteristics in Mobile Crowdsensing Network

Mobile Crowd Sensing (MCS) leverages the widespread use of smart mobile devices equipped with various sensors, making it a powerful tool for collecting and sharing environmental data. Although MCS offers significant advantages in terms of wide coverage, low cost, and real-time information acquisition, it faces major challenges in accurately estimating the truth from the collected noisy and heterogeneous data. However, previous works in truth discovery have primarily relied on either outlier detection or reputation evaluation mechanisms. The outcomes of such methods are not always satisfactory, particularly in scenarios requiring long-term data collection, where the collected data exhibit varying characteristics across different time periods. Therefore, this work addresses above challenge by designing an adaptive truth discovery method to estimate truth value from noisy and heterogenous sensory data. Specifically, our work proposes novel dispersion-based reputation estimation technique, an optimal user set selection strategy and an integrated framework that combines these components to ensure accurate truth discovery across diverse scenarios. Through extensive simulation experiments, the results indicate that our proposed algorithm improves the accuracy of truth discovery and enhances the reliability of the MCS system. Compared to the best baseline in a series of comparative experiments, our proposed algorithm reduces the root mean square deviation (RMSD) of the truth estimation by 68.2% in low dispersion dataset and by 87.3% in high dispersion dataset.