Data Mining Approach to Detection of Random Access Sleeping Cell Failures in Cellular Mobile Networks

Modern cellular mobile networks have become diverse and complex in their nature. On one hand, networks simultaneously operate multiple Radio Access Technologies (RATs), and inside each there several releases which might be used in different geographical areas of the network. On the other hand, cellular systems become more heterogeneous due to growing deployment of femto and pico cells on top of macro layer. A roadmap towards 5G networks implies coexistence of different technologies, various cell sizes, device types, and appearance of new applications and behavior patterns. As a result, quality of provided service and reduction of operational expenditures of the networks are becoming critical aspects of competition between mobile network operators. Earlier, e.g. in 2G networks, it was sufficient to provide good coverage with support of basic call and text message services. Nowadays, users demand for high speed, low delay data transmissions with minimum amount of failures and outages. To comply with the new quality needs operators have to be very efficient in running their networks. This situation poses additional technological challenges for developers of modern mobile networks and researchers in the area of network optimization and intelligent performance monitoring. One of the main accepted approaches designed for network efficiency improvement is the concept of self-organization which relies on automation based on closed loop sensing. This idea has been initially appeared as a set of requirements in Next Generation Mobile Networks (NGMN) alliance [1], [2]. In addition, more thorough description of different use cases of Self-Organizing Network (SON) has been prepared within FP7 SOCRATES project [3]. As a result, 3 Generation Partnership Program (3GPP), involved in development of standards of LTE mobile networks, has included SON use cases to standardization work [4]. There are three categories of SON: selfconfiguration, self-optimization and self-healing [5]. Initial steps of network setup and deployment are covered with selfconfiguration. When network is configured and operates, selfoptimization is used for automatic tuning of network parameters. This functionality is based on performance