An empirical study: Use autoencoder model to determine the abnormal data

Networks play a crucial role in today’s world and failures in network infrastructure can result in significant financial losses and damage to a company’s’ reputation. Predicting future failures can help network administrators take proactive measures to prevent network failures.For dealing with the prediction problem, just focus on the network devices is not enough, we have to focus on the interfaces of the network devices. The Digital Diagnostic Monitoring (DDM) can provide useful information for operation purpose. It contains several measurement value denoted the performance of the interfaces of the network devices.In this survey, we use the real world data to provide the preliminary experiment related to using autoencoder model for predicting future failures in Digital Diagnostic Monitoring (DDM) data. The result of experiment showed that it is better when we set 50th percentile as the reconstruction error threshold to judge the abnormal data. However, it is not acceptable when the cost of dealing with the false positive is a issue.