Multiple linear regression to improve prediction accuracy in WSN data reduction

Carlos Giovanni Nunes de Carvalho (1) , Danielo Gonçalves Gomes (1) , José Neuman de Souza (1) , Nazim Agoulmine (2)

1 GREat - Group of Computer Networks, Software Engineering and Systems
2 IBISC - Informatique, Biologie Intégrative et Systèmes Complexes

Simple linear regression is usually used for WSN data reduction. The mechanism is concerned about energy consumption, but neglects the prediction accuracy. The prediction error from it is often ignored and inconsistencies are forwarded to the user application. This paper proposes to use a method based on multiple linear regression to improve prediction accuracy. The improvement is achieved by multivariate correlation of readings gathered by sensor nodes in field. Tests show that our solution outperforms some current solutions adopted in the literature.

Keywords: multivariate correlation; Prediction accuracy; wireless sensor network; data reduction; linear regression functions

Domains: Networking and Internet Architecture [cs.NI]

Carlos Giovanni Nunes de Carvalho, Danielo Gonçalves Gomes, José Neuman de Souza, Nazim Agoulmine. Multiple linear regression to improve prediction accuracy in WSN data reduction. 7th Latin American Network Operations and Management Symposium (LANOMS 2011), Oct 2011, Quito, Ecuador. pp.1-8, ⟨10.1109/LANOMS.2011.6102268⟩. ⟨hal-00745080⟩