Abnormal Behavior Detection Based on Traffic Pattern Categorization in Mobile Networks

Abnormal behavior in mobile cellular networks can cause network faults and consequent cell outages, a major reason for operational cost increase revenue loss operators. Nonetheless, outages be avoided by monitoring abnormal situations the acting accordingly. Thus, anomaly detection is an important component of self-healing control management. Network operators may use detected to quantify numerically their intensity. The quantification assists characterization potential regions infrastructure updates support creation public policies local connectivity enhancements. We propose unsupervised learning solution using Call Detail Records (CDR) data. evaluate our real CDR data set provided Italian operator compare it against other state-of-the-art solutions, showing performance improvement around 35%. also demonstrate relevance considering distinct traffic patterns diverging geographic areas networks, aspect often ignored literature.