Predicting Network Congestion with Machine Learning

this paper discusses how machine learning methods can improve the complexity management of customer attrition networks. User turnover is insufficient in hybrid cellular networks because losses causing delays and failures due to communication errors are handled. As a result, this research work attempts towards automating a loss of classifier's construction from a database created by random network topology simulations using machine learning techniques. This classifier must fulfil both a computational and a time-varying restriction on its misclassification rate on congestion losses to be useful in this application. With these two constraints in mind, multiple machine learning algorithms are compared. Decision tree boosting tends to be the best option for this application. It surpasses the suggested ad hoc descriptive statistics also in networking research. When combined with customer churn, it increases bandwidth usage dramatically over wireless networks while maintaining good customer churn behaviour over wired networks. This research demonstrates the interest in using computer networks; algorithms are used to build protocols.