An Intelligent Scheme for Big Data Recovery in Internet of Things Based on Multi-Attribute Assistance and Extremely Randomized Trees

Abstract Due to the inherent characteristics of sensor nodes in Internet of Things, such as constrained energy, data redundancy, limited communication range and computing capabilities, the data loss problem becomes one key issue for applications which depends heavily on the data completeness. Some of the current solutions, such as interpolation, are designed to use data correlation for the data recovery problem. Spatiotemporal correlation is an important characteristic of sensory data since the nodes are generally deployed to observed similar physical phenomenon. However, it is very difficult to extract data correlation especially in case that the coupling degree between different perceptual attributes is low. Machine learning is an efficient auto-learning method that can obtain the inherent rules automatically. This paper has proposed an intelligent recovery scheme for big data in Internet of Things based on Multi-Attribute assistance and Extremely randomized Trees (MAET). Firstly, the collected dataset is denoised by detecting and removing the outliers. Secondly, the slave attributes are chosen whose correlations are high with the target attribute by using the Spearman correlation coefficient. Thirdly, the proposed scheme is trained by using extremely randomized trees with slave attributes. Finally, the missing data can be recovered with the trained model as well as the help of other attributes whose data is not lost. Experiment shows that the proposed scheme with multiple attributes is efficient and can improve the accuracy of recovered data compared with other algorithms.