Forecasting internet traffic patterns for the campus Metro-E network using a hybrid machine learning model

Abstract

Complex traffic patterns lead to crucial campus Metro-E network management and resource allocation. This paper presents an internet traffic forecasting by pre-processing data to offer better bandwidth quality of service (QoS). Eight (8) campuses' traffic data were analysed for modelling predictions using statistical analysis. A Metro-E campus network presents four (4) locations: A, E, F, and H have is a strong correlation between inbound and outbound traffic, with correlation values between 0.4547 and 0.5204. As the inbound traffic increases, outbound traffic tends to rise as well. Conversely, locations B, C, and G have weak correlations, indicating more independent traffic patterns. Data outliers were found for locations C and F, where unusual traffic spikes require further network exploration and show key trends in traffic data. Descriptive statistics reveal notable differences, with H has the highest average traffic at about 75 Mbps, while C has the lowest at around 30 Mbps. Location F shows the greatest traffic fluctuation with a standard deviation of 0.4076, whereas Location G has very little fluctuation with a standard deviation of 0.0240. Overall, this pre process data is use to combine machine learning (ML) to improve prediction abilities for better bandwidth management and real-time handling in digital campus environments.