A Correlation Analysis-Based Mobile Core Network KPI Anomaly Detection Method via Ensemble Learning

With the development of new networks, applications in mobile communication networks, such as mobile payment and online classes, have become an indispensable part of people's lives. The core network is one of the most important components of the mobile communication network, which is not only essential but also complex. If the core network fails, it will cause a substantial economic loss. To ensure the reliability and stability of the mobile core network, operators need to detect abnormalities in Key Performance Indicators (KPI,e.g., average response time). Datasets of KPI are usually unbalanced and have a wide range of features. Therefore, we propose a correlation analysis-based KPI anomaly detection via an ensemble learning frame. This frame first performs data augmentation on the dataset using SMOTE oversampling algorithm and uses the Pearson correlation coefficient method for feature selection, then construct an ensemble learning XGBoost-based anomaly detection method for KPI. Finally, we evaluate our scheme with the confusion matrix. The results show that our scheme obtained a high accuracy and recall rate. The training and testing dataset we collected is a KPI dataset of a Chinese operator for the first three months of 2020. It is worth noting that no relevant studies used this dataset before.