Machine Learning for Network Supervision and Fault Prediction in telecommunication infrastructures

In recent years, Machine Learning algorithms spread consistently in a lot of fields thanks to their strong potential and ability in dealing with relevant and common problems such as classification, prediction, regression, pattern analysis etc. The main purpose of this project is to exploit Machine Learning solutions for fault management and network monitoring in telecommunication infrastructures. Monitoring systems are usually used, by telecommunication companies, to detect and record the malfunctioning during the journey of the signal from the service provider to costumers, in particular, this work is focused on what happens when the signal arrives in the cabinets, devices positioned along the roads that make the signal travel in the last kilometres. The project involves data from physical network infrastructures collected during more than one year, this data represents critical events notified as alarms and described by both autonomous devices and humans who work close to the infrastructure. The sequences of alarms and the requests of intervention needed to resolve potential failures, that are detected and recorded by monitoring systems, are the core data of the analysis. The aim is, therefore, to explore several algorithms to analyse the sequence of alarms, understand their severity, detect recurrent patterns, and predict the circumstances for which new alarms and intervention requests may occur. In this regard, Pattern Mining techniques and Machine Learning algorithms with Gradient Boosting methods for Classification, together to frameworks for Big Data management, like Apache Spark and Hadoop, are the main actors of this project.