Discovering Alarm Correlation Rules for Network Fault Management (video)

In this blog post, I will share the video of our new paper about analyzing alarms in telecomunication networks presented at the AIOPS 2020 workshop. This work is part of an industrial collaboration project. The motivation for this project is that there are typically thousands of alarms in a telecomunication network, and not all of them are important. To allows network operators to focus on fixing issues that are the most important, we propose a method to discover correlations between alarms.

For this purpose, we view a telecommunication network as an attributed graph where nodes represent devices, edges indicates connections between devices, and attributes of vertices represent alarms. Then, we apply a novel algorithm to find rules of the form A–>B indicating that if alarm A appears, Alarm B is likely to occur. Then, using these rules, we can reduce the number of alarms presented to network maintenance workers. Though, the approach is designed for analyzing alarms it could be applied to other data modelled as graphs.

Here is the link to watch the paper presentation:

And here is the reference to the paper:

Fournier-Viger, P., Ganghuan, H., Zhou, M., Nouioua1, M., Liu, J. (2020). Discovering Alarm Correlation Rules for Network Fault Management. Proc. of the International Workshop on Artificial Intelligence for IT Operations (AIOPS), in conjunctions with the 18th International Conference on Service-Oriented Computing (ICSOC2020) conference,

That is all I wanted to write for today!

Philippe Fournier-Viger is a full professor working in China and founder of the SPMF open source data mining software.

This entry was posted in Data Mining, Data science, Video and tagged , , , , . Bookmark the permalink.

5 Responses to Discovering Alarm Correlation Rules for Network Fault Management (video)

  1. Yulai Cui says:

    Thanks for the insightful paper. How should I infer rules if ACORa2b = ACORb2a = 1? In other words, when all A happen with B?


  2. Mohsen Keshavarzi says:

    thanks for sharing.
    i have one question.what is the best algorithm to do this job with ML?

    • Hi, thanks for reading and sorry to reply late. I did not see your comment before. For ML, I think that pattern mining can be viewed as a ML technique. But I think maybe you want to know about something more like neural networks. I do not know the answer. 🙂 But by reading some papers we could find out. Best regards,

  3. Ben Julio says:

    In the video you mentions that it can be implemented using “Pattern Mining” or ML. Isn’t that pattern mining or pattern recognition is a type of ML?

    • Hi, pattern mining could be viewed as a form of ML but I think it depends. Let me explain my perspective.

      As several persons, I define machine learning as a computer programs (algorithm) that can learn to do a task by doing it or by learning it from some data.

      In the case of pattern mining, the goal is to analyze data rather than to do a task. A pattern mining algorithm will find patterns in the data that may help a human to understand the data. For example, analyzing shopping data, we may find a pattern that many customers buy bread and wine together. This kind of pattern may help a human to understand the behavior of the customers, but the pattern mining algorithm is not learning to do a task. Pattern mining can thus be viewed as a technique to explore the data… The pattern mining algorithms just extract the patterns that meet the requirement of the user. For this reason, I prefer to think about pattern mining algorithms as data mining techniques rather than machine learning techniques. However, it is possible to use the patterns found by a pattern mining algorithm to do prediction or classification. In this case, if we add this step of using the patterns to do a task such as prediction of classification, then we can consider that this is machine learning.

      Also, pattern mining should not be confused with pattern recognition which is a different field of research. Pattern recognition typically focus on processing images, videos or audio data. For example, one may train a neural network to do a task such as recognizing cats in pictures. In this case, this is machine learning because we teach the model how to perform a task.

      In pattern mining, it is different. We want to extract a well-defined type of pattern that is known before hand and the algorithm will just enumerate all the patterns that match the constraints.. There is no task to learn!
      Best regards,

Leave a Reply

Your email address will not be published. Required fields are marked *