Wireless sensor networks allow fine-grained observations of real-world 
phenomena. However, providing constant measurement updates incurs high 
communication costs for each individual node, resulting in increased energy 
depletion in the network. Data reduction strategies aim at reducing the amount 
of data sent by each node, for example by predicting the measured values both 
at the source and the sink node, thus only requiring nodes to send the 
readings that deviate from the prediction. While effectively reducing power 
consumption, such techniques so far needed to rely on a-priori knowledge to 
correctly model the expected values. Our approach instead employs an algorithm 
that requires no prior modeling, allowing nodes to work independently and 
without using global model parameters. Using the LeastMean-Square (LMS) 
adaptive algorithm on a publicly available, real-world (office environment) 
temperature data set, we have been able to achieve up to 92% communication 
reduction while maintaining a minimal accuracy of 0.5 degree Celsius.