Educational Network Traffic Data Analysis

This research work focuses on looking into the educational network data traffic with the view to understand current trends in order to plan. The purpose of the study is to monitor the London South Bank University (LSBU) network traffic, and to develop a software analytic tool that can analyse the network traffics. The outcomes of this case can help organisations to enhance their network performance, and provide recommendations for scalability and future upgrades to meet increasing traffic volumes and satisfy user expectations in terms of quality of service (QoS). For this case, LSBU network traffic raw data is first captured using the Paessler Router Traffic Grapher (PRTG) network-monitoring tool, and then numerical analysis algorithms are developed to examine the captured raw data. A fast Fourier transform (FFT) algorithm is developed to study the traffic’s frequency domain information, performing large-scale analytics for user behaviour and initial results of the FFT show that it has potential as a tool for measuring and/or predicting LSBU behaviour. A wavelet-based algorithm is also used in the development in order to decompose, denoise, and compress the data, as well as analyse the time frequency characteristics of the data. In addition, nonlinear autoregressive exogenous model (NARX) based on the Levenberg-Marquardt backpropagation algorithm technique is applied to study and predict data usage in its current and future states, as well as visualise the hourly, daily, weekly, monthly, and quarterly activities with less computation requirement. Results and analysis prove the accuracy of the prediction techniques. The research shows a dynamic Neuro-fuzzy local modelling system (DNFLMS) for calling the dynamic network data traffic in time-step ahead, by building up the forecasting models to predict the bandwidth data affected by the network traffic, from current time to one day, three days, one week, two weeks and one month ahead dynamically at different time intervals. This research has shown that the proposed DNFLMS is superior in terms of both model performance and computational efficiency to those models that adopt a batch-learning algorithm such as a multi-layer perceptron (MLP) system trained using the back-prorogation learning algorithm (MLP-BP)