Attempt to Apply Machine Learning to a Failure Database - A Case Study on Communications Networks

Progress in IT has resulted in great improvements in convenience. However, IT can cause failures that have significant negative impacts such as system failures. In order to improve these circumstances, it is important to accumulate and analyze numerous past failure cases. In order to achieve this purpose, the authors have applied machine learning to a previously accumulated failure database. We have constructed a mechanism by which to calculate the degree of similarity between documents by two methods. One method uses the appearance frequency of words, and the second method uses the appearance probability of each topic extracted from the whole document. In the present paper, focusing on communications network failures, we realized a function by which to extract past failure cases similar to inquiry inputs, as new failures. A detailed analysis and comparison of these results extracted by these two methods are presented.