Data recovery in wireless sensor networks based on attribute correlation and extremely randomized trees

In wireless sensor networks, collected data usually have a certain degree of loss and are unable to meet actual application needs due to node failures or energy limitation, etc. The current data recovery methods in wireless sensor networks focus on the usage of spatial-temporal correlation between perceptual data but seldom exploit the correlation between different attributes. This paper proposes a data recovery algorithm based on the Attribute Correlation and Extremely randomized Trees (ACET). Firstly, the Spearman's correlation coefficient is adopted to construct the correlation model between different attributes. In case that a given attribute is lost, the correlation model is used to select other attributes that have a strong correlation with this attribute, and then take advantage of them to train the extremely randomized trees. Finally, the lost data can be recovered by the trained model. Experimental results show that the correlation between attributes can improve the effectiveness of data recovery compared with other methods.