Sleep-scheduling mechanisms are often used in wireless sensor networks (WSN), as they can save a significant amount of energy by turning off the redundant nodes in the network. In a recent survey about energy-efficient scheduling mechanism in WSNs it was however shown that there are virtually no protocols that support a mobile environment. In this paper we propose thus an adaptive energy-efficient sleep scheduling mechanism that is able to handle a dynamic environment, mobile nodes and moving events. The proposed solution is fully distributed, can cope with frequent node failures, and balances energy consumptions, which can significantly extend the global network lifetime. The solution is adaptive, since it can dynamically detect and learn the existence of a linear relation between sensor measurements, eliminate the redundancy, and estimate the deficient data based on the learned relations. The extrapolation error can be well controlled by fine-tuning the threshold of expected inaccuracy, which is a user-specified parameter. We compare our solution with deterministic clustering approaches, provide parameter sensitivity analysis and discuss the simulation results.