Abstract: A large amount of system log data is collected during network operations, and finding out precise system faults has become an important research direction. This paper proposes a conditional causality mining algorithm (CCMA), which generates a set of time series data from log messages, and uses Fourier analysis and linear regression analysis to delete a large number of irrelevant periodic time series. Then, the causal inference algorithm is used to output the directed acyclic graph, and the final result is obtained by detecting the edge distribution of the acyclic graph and eliminating the redundant relationship. The simulation results show that the CCMA algorithm outperforms the dependent mining algorithm (DMA) and the network information correlation and exploration algorithm (NICE) in two main performance indicators such as processing time and edge correlation rate, which proves that the CCMA algorithm can effectively optimize the processing speed and mining accuracy in log event mining.

Key words: conditional causality, log data, network management, data mining