Network traffic data recovery based on local tensor completion

• 1、Department of Electrical and Computer Engineeering, Hunan University, Changsha 410000

Abstract：As the network scale continues to expand, it is unrealistic to monitor the traffic data of all nodes in the network. Inferring the overall network traffic data from some measurement data becomes one of the more and more important tasks in the project. With multidimensional data structure features, tensor completion is a promising technique to infer lost data more accurately. A novel local tensor completion model is proposed in this paper. The data recovery accuracy is improved by using local data to form and recover each of the local tensor with a low rank structure. In order to overcome the challenges in the research, several new techniques are proposed, including efficient calculation of candidate anchors based on local sensitive hash (LSH), selection of appropriate anchor points to construct sub-tensors, and coding of factor matrices to facilitate data loss. The local tensor was found with similar elements, as well as local tensor completion and data fusion . Finally, experiments were carried out using the real data sets. The experimental results show that the relative error of the sampled data and the error of infered data are compared with the classical tensor balance factor (CP) decomposition in the case of the same sampling rate and rank, reduced by 7\% and 3\% respectively. The research results of this paper can be used to recover network traffic data more accurately.

Keywords： Tensor completion Network measurement data Local sensitive hash