To conduct necessary aggregation on varying-quality sensed data uploaded by workers in mobile crowdsensing, truth discovery technology has emerged as the cornerstone for providing precise data support for subsequent applications. Existing studies tend to adopt local differential privacy for protection against potential privacy breaches, but often ignore the influence of outliers in the sensed data on the truth discovery accuracy under local differential privacy. These outliers often have a large range of values, resulting in a large amount of noise in the injected data. Additionally, due to workers’ concerns about privacy breaches, mobile crowdsensing servers cannot preprocess data without privacy protection. To this end, this study proposes NATURE, which meets local differential privacy based on adaptive pruning. The core idea of the algorithm is to consider the noise types in the data to adaptively prune all unnecessary workers’ values or certain task values. In NATURE, the noise-aware weight and importance estimation (NWIE) method based on a formalized constraint optimization problem is designed to facilitate data pruning. Based on proving the optimal pruning problem is NP-hard, this study designs the utility-aware adaptive pruning (UAP) method with polynomial time complexity to conduct pruning. Furthermore, a theoretical analysis of NATURE’s privacy, utility, and complexity is carried out. Experimental results on two real-world datasets and one synthetic dataset demonstrate that NATURE achieves an accuracy improvement of at least 20% in obtaining “truth” compared to its comparative algorithms.