Propuesta de algoritmo de control de congestión en redes de paquetes usando técnicas de machine learning

Abstract

Abstract: In this study, the challenge of congestion control in packet networks is addressed through the application of advanced machine learning techniques. Using network topologies designed in OMNeT and a detailed dataset that includes key metrics such as capacity, datarate, utilization, queueing time, and rates of sent, received, and lost packets, a predictive model is developed. This model, evaluated using machine learning tools such as Random Forest and SVM in MATLAB, demonstrates notable accuracy (96% with Random Forest, 93% with SVM and Naïve Bayes 91%) in predicting network congestion. Specific criteria are established to identify congestion, such as utilization over 98%, queueing time greater than 1 ms, and a packet loss rate exceeding 2%. The application of cross-validation reinforces the reliability of the models. This approach not only offers an effective method for predicting congestion but also sets a precedent in network management, combining advanced data analysis with machine learning techniques.