Automated anomaly detection in network management

Analyzing patterns from information stream and searching for anomalies will
reveal the surprising in any quite business. For applications involving many information
streams, the challenge of detecting anomalies has become tougher over time, as information
will dynamically evolve in delicate ways in which following changes within the underlying
infrastructure. As datasets increment in size and unpredictability, the human effort expected to
look at dashboards or keep up rules for perceiving framework issues or business problems ends
up impractical. The automated detection of potential business process anomalies could
colossally help the business and different process members identify and comprehend the
reasons for process errors.

In this project, an automated anomaly detector system is created using Robust
Principal Component Analysis (RPCA) which identifies a low rank representation, random
noise, and a set of outliers by repeatedly calculating the SVD and applying thresholds to the
singular values and error for each iteration. In the network management system of the company,
certain issue trackers called alarms arises whenever a network element is disturbed. In the
current system, the alarms are handled based on the threshold in the network graph based on
logs and inputs. Whenever a there is a spike or dip in the graph, they are marked as alarms and
categorized based on the level of point they surpassed. In the proposed system, this technique
will be replaced by a more enhanced and optimized technique which uses machine learning to
find outliers by analyzing the log data received.

MSC MI 2016-2018 INT Dr. Asharaf S