Short-Term Traffic Forecasting in Optical Network using Linear Discriminant Analysis Machine Learning Classifier

Efficient resource allocation is a key task for network operators to provide services for users. Its optimization, in short-term perspective, allows to reduce operating costs, provide required quality of service and detect dataflow anomalies. It is crucial to develop tools to better understand and forecast traffic behaviour of optical networks. In this work, we propose a Machine Learning (ML) procedure for short-term traffic volumes forecasting, using the Linear Discriminant Analysis (LDA) classifier. The main novelty, comparing to the other works in this field, is that in our approach, we predict fixed bitrates levels of the traffic, instead of exact traffic volume. In consequence, the traffic prediction problem is formulated as a classification task, while most of the previous works in this field model the traffic using time series. We also present numerical results as a proof of effectiveness of the described approach. We examine real traffic collected by Seattle Exchange Point, together with traffic generated based on real data (which consist of real traffic dependencies). We obtained up to 93% of correct predictions for real data traffic.