ACCURATE INFERENCE OF UNSEEN COMBINATIONS OF MULTIPLE ROOTCAUSES WITH CLASSIFIER ENSEMBLE

Root cause analysis (RCA) of network faults is crucial to wireless network operation and management. It, however, is challenging, due to diverse feature types, diverse lengths of time slices, simultaneous occurrences of multiple root causes, and lack of training samples. In this paper, we present our solutions for these problems in ICASSP-SPGC-2022 AIOps Challenge in Communication Networks. We first design specific feature engineering method to represent the provided spatial features. Secondly, we conduct time series analysis on training data and propose an efficient method to infer whether a sample includes multiple root causes. Thirdly, in order to solve the lack of training data for unseen combinations of multiple root causes, we propose to ensemble multiple single-root-cause classifiers. Fourthly, we introduce TextCNN into multivariate time series classification to obtain high accuracy. Our approach achieved score 0.93 and ranked 4th place on the leader board.