Evaluation of Machine Learning Algorithms for Automated Management of Wireless Links.

Machine learning, a subfield of artificial intelligence, has been widely used to automate tasks usually performed by humans. Some applications of these techniques are understanding network traffic behavior, predicting it, classifying it, fixing its faults, identifying malware applications, and preventing deliberate attacks. The goal of this work is to use machine learning algorithms to classify, in separate procedures, the errors of the network, their causes, and possible fixes. Our application case considers the WiBACK wireless system, from which we also obtained the data logs used to produce this paper. WiBACK is a collection of software and hardware with auto-configuration and self-management capabilities, designed to reduce CAPEX and OPEX costs. A principal components analysis is performed, followed by the application of decision trees, k nearest neighbors, and support vector machines. A comparison between the results obtained by the algorithms trained with the original data sets, balanced data sets, and the principal components data is performed. We achieve weighted F1-score between 0.93 and 0.99 with the balanced data, 0.88 and 0.96 with the original unbalanced data, and 0.81 and 0.89 with the Principal Components Analysis.