TreeBased DataAggregation inSensor Networks Using Polynoial Regression
Torsha Banerjee, Kaushik Roy Chowdhury, DharmaP . Agrawal · 2005
Inthis paper, wepropose atree basedregression algorithm, (TREG)thataddresses theproblemofdata compression in wireless sensornetworks. By function approximation based onmultivariable polynomial regression and passing onlythecoefficients returned bytheregression function instead ofaggregated data, TREGachieves thefollowing goals: (1)thesinkcangetattribute values inregions devoid ofsensor nodes forattribute values that showsmooth spatial gradation (2) readings overanyportion oftheregion canbeobtained atone timebyquerying therootinstead offlooding those regions, thus incurring significant energy savings. Assize ofthedatapacket transmitted fromonetreenodetoanother remains constant, the proposed schemescales wellwithgrowing network density. Extensive simulations areperformed on realworlddatato demonstrate theeffectiveness ofouraggregation algorithm. Results reveal that foranetwork density of0.0025, theoptimal tree-depth should be4inorder torestrict theabsolute error to less thanathreshold of6%.A datacompression ratio ofabout 0.02isachieved using ourproposed algorithm, whichisalmost independent oftree depth. IndexTerms-Attribute-based Trees, DataAggregation, Function-approximation,