Prediction and influence factors analysis of IP backbone network traffic based on Prophet model and variance reduction

Accurate and efficient traffic prediction directly determines the construction scale and investment budget of communication networks, which is crucial for network planning. Despite the rise of popular machine learning models, traditional statistical models maintain significant advantages in interpretability, controllability and simplicity, retaining an essential role in contemporary communication network traffic prediction. This paper analyzes and predicts the inter-provincial egress traffic of 31 provinces in a large-scale operational IP backbone network using traditional regression analysis, the time series Prophet model, and a novel combination of these two prediction models. We explore the applicability of these prediction methods for inter-provincial egress traffic. Additionally, we systematically study the interactions between eight types of macroeconomic factors and the inter-provincial egress traffic of the IP backbone network. Our statistical results indicate varying degrees of correlation between inter-provincial egress traffic and five types of social macroeconomic indices, as well as three types of communication industry indices. Notably, four indices-Gross Domestic Product (GDP), per capita disposable income, per capita consumption expenditure, and the number of Internet broadband users-exhibit high correlation. Among the four forecasting models constructed, the overall forecasting effectiveness is ranked from best to worst: time series Prophet combined with variance reduction model, time series Prophet model, correlation regression analysis model, and stepwise regression model. Our innovative combined model improves the average prediction accuracy of the Prophet model by 2.6 %, achieving 100 % effective prediction of traffic in all 31 provinces with an overall excellence ratio of 94 %. This approach is highly applicable and effective, gaining popularity in engineering practice and large-scale application.