Usage of control charts to fraud detection in the voice traffic transit

The continuous evolution of fraudulent telecommunication traffic necessitates the improvement of anti-fraud systems designed to prevent spam and fraud. The anomaly detection module in such systems utilizes control charts for monitoring, which are plotted based on time series of voice traffic transmission characteristics. These time series are non-stationary, cyclic, and asymmetrically distributed, complicating their analysis. This paper presents an algorithm that reduces the frequency of false positives in control charts and enhances the overall efficiency of anti-fraud systems. The algorithm involves preparing the original time series of voice traffic transmission characteristics for control chart analysis by segmenting the non-stationary series with a cyclic component and subsequently merging homogeneous segments into a single sample.