Missing data is unavoidable in sensor networks due to sensor faults, communication malfunctioning and malicious attacks. There is a very little insight in missing data causes and statistical and pattern properties of missing data in collected data streams. To address this problem, we utilize interacting-particle model that takes into account both patterns of missing data at individual sensor data streams as well as the correlation between occurrence of missing data at other sensor data streams. The model can be used in algorithms and protocols for energy efficient data collection and other tasks in presence of missing data.

We use statistical intersensor models for predicting the readings of different sensors. As a driver application, we address the problem of energy efficient sensing by adaptively coordinating the sleep schedules of sensor nodes while we guarantee that values of nodes in the sleep mode can be recovered from the awake nodes within a user’s specified error bound and probability of missing data at awake nodes is less than a given threshold. The sleeping coordination is addressed by creating the maximal number of subgroups of disjoint nodes, each of whose data is sufficient to recover the data of the entire network in presence of missing data. On simulated and actually collected data for temperature and humidity sensors in Intel Berkeley Lab, we show that by using sleeping coordination that considers missing data, we reduce the typical 40% missing data rate of traditional sleeping techniques to less than 7%.