Abstract: In wireless sensor networks that consist of a large number of low-cost, battery-powered sensors, one of the main challenges is to obtain long system lifetime without sacrifying quality of service such as sensing coverage and data integrity. Scheduling sensors to work alternatively can prolong lifetime efficiently. In this paper, a novel data-driven sleeping scheduling mechanism is proposed, which can extend lifetime by identifying redundant nodes based on time-spatial correlations among sensing data. The main idea is: first, a non-parametric regression method is exploited to develop prediction models for forecasting measurements of one sensor using data from other sensors; then the maximal number of node dominating sets is created; finally the sleep?duty cycles of these node dominating sets based on prediction models are scheduled. Data in each of the dominating set is sufficient to recover the measurements of the entire sensor network. We present the centralized, semi-distributed and distributed sleeping scheduling algorithm respectively, guaranteeing that values of sleeping nodes can be recovered from awake nodes within a user's specified error bound. It is known that this is the first work on data-driven sleeping scheduling for large scale sensor networks. Experiments results show that the proposed methods can prolong network lifetime substantially while maintaining data integrity under the user's error constraint.