Completion of Traffic Matrix by Tensor Nuclear Norm Minus Frobenius Norm Minimization and Time Slicing

The goal of traffic matrix completion is to estimate the elements of the network traffic matrix that could not be observed. Conventional methods can be divided into mathematical optimization-based methods and machine learning-based methods, with the former suffering from low prediction performance and the latter depending on the dataset and missing patterns used for training. We propose an optimization-based traffic matrix completion method with high accuracy that uses the nuclear norm minus Frobenius norm, which has recently been shown to be effective in image denoising. The proposed method combines the construction of a tensor by dividing the traffic matrix by a predetermined time interval and the introducing of regularization by a newly defined tensor nuclear norm minus Frobenius norm. Experimental results show that the proposed method achieves traffic matrix completion comparable to or better than state-of-the-art neural network-based methods, without the need for prior training.