Path Planning Optimization in SDN Using Machine Learning Techniques

• Published in: 2021 IEEE Fifth Ecuador Technical Chapters Meeting (ETCM) pp. 1 - 6
• Main Authors: Rodriguez, Marlon, Moyano, Ricardo Flores, Perez, Noel, Riofrio, Daniel, Benitez, Diego
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12.10.2021
• Subjects: ANN; Costs; kNN; Machine learning; Path planning; Quality of service; SDN; Support vector machines; SVM; Telematics; Traffic control
• DOI: 10.1109/ETCM53643.2021.9590749

Internet, mobile networks, and mobile devices have contributed to the massive development of telematics applications. Therefore, the underlying communication network that supports the connectivity of these applications must provide an adequate level of QoS. On the other hand, the advent of new networking paradigms such as Software Defined Networks (SDN) has transformed the telco landscape. Consequently, traditional teletraffic engineering techniques cannot comply with the requirements of agile, dynamic, and tailored traffic controls. In this context, a proposal to improve the QoS of communication networks by optimizing the path planning process using the machine learning principles is presented. Thus, path planning is considered a multi-classification problem. Several configurations of three machine learning classifiers have been evaluated to determine the best model. Two class-balanced experimental datasets named Dl and D2 were created for validation purposes. The support vector machine classifier with a linear kernel and cost c = 10 2 was the best model obtaining a mean of an area under the receiver operating characteristics curve of 0.999 using the D1 dataset. The same classifier with a polynomial kernel and cost c = 10 achieved a score of 0.999 using the D2 dataset. These results statistically overcame the remaining classification schemes at a = 0.05, determining the support vector machine model as the best classifier to find optimal paths between endpoints.