I have a sequence of logs with time stamps in json format which needs to get classified into success and failure(failure class with suitable reasoning like why it is failing). Here the logs does'nt have labels so it is obvious to use an unsupervised model.The classification should focus on relating two or more logs to find the failure, as failure case can be detected only by analysing a group of instances.

I have already tried K-means clustering where it is fine to get the outputs but unable to figure it out whether they are correct or wrong as it has no labels to calculate the accuracy and this clustering does'nt involve correlation in the dataset(i.e analysis of two or more logs to find the failure)

Actual output is binaries(0 or 1) but i expect the output to be a meaningful reason that why it failed

0 Answers