Active Learning-Based Fault Diagnosis in Self-Organizing Cellular Networks | IEEE Journals & Magazine | IEEE Xplore

Active Learning-Based Fault Diagnosis in Self-Organizing Cellular Networks


Abstract:

Fault diagnosis in self-organizing cellular networks (SONs) is usually considered as a multi-classification problem and can be handled by machine learning methods. Howeve...Show More

Abstract:

Fault diagnosis in self-organizing cellular networks (SONs) is usually considered as a multi-classification problem and can be handled by machine learning methods. However, the main obstacle for practical applications of these methods is the high cost and difficulty of obtaining sufficient labeled data (i.e., statistical data with labels which clearly state the fault cause). To address this issue, we propose a novel active learning based fault diagnosis scheme, which can achieve high diagnosis performance with only few labeled training instances and thus significantly lowers the cost. The core idea is to select the most valuable unlabeled data for labeling and training. The process of selection is completed by uncertainty sampling. Experimental results, based on an LTE network fault model database, reveal that to achieve the same diagnosis accuracy (e.g., 99%), the proposed scheme requires significantly less labeled training instances comparing with existing non-active methods (e.g., 66 vs 557). Also, under the same number of labeled instances, the proposed scheme can achieve higher performance under three different metrics.
Published in: IEEE Communications Letters ( Volume: 24, Issue: 8, August 2020)
Page(s): 1734 - 1737
Date of Publication: 30 April 2020

ISSN Information:

Funding Agency:


References

References is not available for this document.