A CNN-based network failure prediction method with logs

Nowadays people are becoming increasingly dependent on networks. Taking precautions before failure of networks is becoming more and more important. Compared with detection of faults, prediction of faults is more critical as it can efficiently avoid the damage made by network faults based on the fault prediction. So this paper focuses on network failure prediction. When predicting network faults, current methods mainly rely on structured data, like alarm data. Compared with structured data, log data are more abundant and recently implemented to solve failure prediction problems of network systems. In order to predict failure in the networks, it becomes essential to take advantage of log data. In this paper, we propose a novel method to predict failure in the network system. We build a CNN model to accomplish our work. Experimental results confirm the effectiveness of our method.