Applying machine learning methods to classify wireless link errors, their causes and solutions
Wireless links are often prone to performance degradation due to signal diffraction, interference from other sources or even radars. To identify the various causes when they occur in an operational network, we have analysed the involved MAC protocols and data forwarding mechanisms, and studied an extensive amount of logging information to derive a set of monitoring data. We then investigate if and which machine learning algorithms can be applied to automate the processing of the monitoring data to accurately determine the cause of a link degradation in near real-time. We train and test neural networks, decision trees, support vector machines, and K-nearest neighbours. We also describe the process of data synchronisation, data set creation, and the numerical variable selection for the classification of link degradation causes. Our findings show that decision trees achieve the highest accuracy, while neural network and support vector machines also achieve high performance.