Title: MBAN-MLC: a multi-label classification method and its application in automating fault diagnosis

Authors: Shijun Chen; Lin Gao; Guoqiong Liao

Addresses: School of Computer Science and Technology, Xidian University, Xi'an, China; Corporation Technology, Siemens AG, Beijing, China ' School of Computer Science and Technology, Xidian University, Xi'an, China ' School of Information Technology, Jiangxi University of Finance and Economics, Nanchang, China

Abstract: With increasing competition in the mobile telecom network market, the network quality becomes one of the key factors for competitions between network operators. At the work site of drive-testing in the mobile radio communication network, drive-testing experts needed to tag the fault causes manually based on their experience for fault diagnosis, and there were always multiple causes. Traditional single-label classification method cannot be used here to automatically tag the multiple fault causes. In this paper such kind of fault diagnosis problem is transformed to the multi-label classification problem, and a multi-label classification method (MBAN-MLC) is proposed for fault diagnosis automation. MBAN-MLC is based on a Bayesian network-augmented naïve Bayes model with multiple classifying nodes. In the MBAN-MLC method the relationship between labels are taken into considered to improve the classification precision during the model construction and inference. The MBAN-MLC method is also verified to be effective in the proprietary drive-testing fault diagnosis dataset and standard multi-label dataset, and does improve the efficiency of fault diagnosis of drive-testing greatly in contrast to traditional manual mode.

Keywords: multi-label classification; multi-label learning; Bayesian network; Bayesian network-augmented naïve Bayes; multiple classifying nodes; fault diagnosis.

DOI: 10.1504/IJIMS.2018.095240

International Journal of Internet Manufacturing and Services, 2018 Vol.5 No.4, pp.350 - 364

Available online: 28 Aug 2018 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article