A Generative Adversarial Learning-Based Approach for Cell Outage Detection in Self-Organizing Cellular Networks

For enabling automatic deployment and management of cellular networks, Self-Organizing Network (SON) was boosted to enhance network performance, to improve service quality, and to reduce operational and capital expenditure. Cell outage detection is an essential functionality of SON to autonomously detect cells that fail to provide services, due to either software or hardware faults. Machine learning represents an effective tool for such a task. However, traditional classification algorithms for cell outage detection are likely to construct a biased classifier when training samples in one class significantly outnumber other classes. To counter this problem, in this letter, we present a novel method that is able to learn from imbalanced cell outage data in cellular networks, through combining Generative Adversarial Network (GAN) and Adaboost. Specifically, the proposed approach utilizes GAN to change distribution of imbalanced dataset by synthesizing more samples for minority class, and then uses Adaboost to classify the calibrated dataset. Experimental results show significant improvement of classification performance for imbalanced cell outage data, on the basis of several metrics including Receiver Operating Characteristic (ROC), precision, recall rate, and F-value.