Informer-Based Anomaly Detection in Mobile Networks Using CDR Time-Series Analysis

This paper introduces an Informer model for anomaly detection in mobile networks. Call detail records, containing mobile internet traffic data are thoroughly processed and feature engineering is applied to improve the model's overall accuracy and efficiency. Moreover, logarithmic transformation is employed onto the data to ensure realistic and precise results. After training the model for the length of 5 epochs and testing with synthetic injected anomalous data patterns an accuracy of 80.69% was achieved.