首页
>
资源共享 > Telecommunication Alarm Dataset Repository
Telecommunication Alarm Dataset Repository
Alarm correlation analysis is an important field which can be applied to Alarm Reduction, Alarm Compression, finding the root cause of fault and etc. From the view of data mining, how to find out all of the confident patterns from the alarm datasets is one of the key problems.
Alarms in telecommunication networks are massive, bursting and intermittent. Alarm datasets include the alarms from many millions of network elements of many different types. Here are some alarm datasets which can be downloaded freely to study the confident alarm associate patterns discovery. We have deleted the semantic infomation and the network topology information to simplify the problem.
The alarm datasets are binary files, each type of alarms has been identified a number and represents some kind of alarm from some equipment. Each record in the alarm datasets is 12 Bytes. The first 4 bytes presents the alarm id.The second 4 bytes presents the time (absolute second) when the alarm occurs, the last 4 bytes is reserved and meaningless.
Alarm-a dataset is from a network management system (NMS) of an anonymous operation provider, which includes 6 months/2754641 alarm records of telecommunication network.
Alarm-b dataset is from a network management system of experimental telecommunication network. There is about 15 days alarm data.
Alarm-c dataset involves 4 weeks alarm data from the same NMS as Alarm-a with a diffrent collection duration.To facilitate mining, we have converted these binary datasets to text transaction dataset by setting window parameter and sliding step parameter according to the paper written by H. Mannila. The converted transaction datasets are described as follows. The data format is:
1 23 45 67 234
23 45 1234 167 4 334
12 32 34 33 3 433
...
Alarm1 dataset is from Alarm-a (window=300s and step=150s), which contains 104823 transactions and 96801 alarm id. The maximum length of the transactions is 1062, the minimum length is 1 and the average length is 28.6.
Alarm2 dataset is from Alarm-b (window=120s and step=60s), which contains 8261 transactions and 22255 alarm id. The maximum length of the transactions is 320, the minimum length is 1 and the average length is 29.5.
Alarm3 dataset is from Alarm-c (window=300s and step=150s), which contains 15521 transactions and 11430 alarm id. The maximum length of the transactions is 163, the minimum length is 1 and the average length is 16.9.
Alarm4 dataset is from Alarm-c (window=120s and step=60s), which contains 38699 transactions and 11430 alarm id. The maximum length of the transactions is 140, the minimum length is 1 and the average length is 8.4.Here is a program, convertToTrans.exe, (for windows) to convert a sequence dataset to a transaction dataset. You can try the parameters you want.
if any question, please email to xiaobo, thanks!
xiaobo 资源共享 dataset 136 views
