Multivariate Stream Data Reduction in Sensor Network Applications

We evaluated several multivariate stream data reduction techniques that can be used in sensor network applications. The evaluated techniques include Wavelet-based methods, sampling, hierarchical clustering, and singular value decomposition (SVD). We tested the reduction methods over the range of different parameters including data reduction rate, data types, number of dimensions and data window size of the input stream. Both real and synthetic time series data were used for the evaluation. The results of experiments suggested that the reduction techniques should be evaluated in the context of applications, as different applications generate different types of data and that has a substantial impact on the performance of different reduction methods. The findings reported in this paper can serve as a useful guideline for sensor network design and construction.