Research on Anomaly Detection Method for Telecom Services Based on Improved Autoencoder and Generative Adversarial Network

In the telecommunications industry, abnormal data caused by systems or services can lead to customer churn, hindering the development of telecom services. To address the collection and analysis of such abnormal data, our proposed a framework based on Autoencoder (AE) and Conditional Generative Adversarial Network (CGAN). Firstly, multiple service data tables are collected and associated using the Spark to obtain the dataset. Secondly, the autoencoder reconstructs the original data, calculates the reconstruction error, and extracts the potential features of the reduced dimensionality sequence. Then, the generative adversarial network performs anomaly detection and records the reconstruction error. Finally, we adopt the idea of ensemble learning and combine the reconstruction errors of the autoencoder and the adversarial generative network as the anomaly discrimination criterion. Experiments demonstrate that the proposed AE-CGAN performs well in telecommunications data anomaly detection, accurately identifying data anomalies.