A new approach for multi-source data prediction in Wireless Sensor Networks: Collaborative filtering

The prime shortcoming of Wireless Sensor Networks (WSNs) is their energy constraint. The main energy consumer in a sensor node is its radio transmitter. One of the most effective methods to reduce the data transmission rate is data prediction. By data prediction, the amount of transmitted data is reduced; which results in energy saving and the longevity of the network life. Environmental variations almost have similar effects on different sensor sources in a sensor device. So, considering the correlation between different sources reduces the noise impact and increases data prediction accuracy. In this paper, temporal and multi-source correlations are used, to reduce data transmission in WSNs. We have used item-based collaborative filtering for extracting the relationship between different phenomena sensed by sensors in consequent time points. The extracted information is used to predict data value for the next time points. We conducted our simulations on the actual data collected from 54 sensors deployed in the Intel Berkeley Research lab. According to the simulation results, collaborative filtering reduces transmission rate and computational cost, in comparison to the other state of the art methods. When the error threshold is greater than 0.5, it can decrease more than 98% of data transmissions.