Imbalanced High-Frequency Number Classification Based on DSUS

High-frequency numbers refer to phone numbers that exceed a certain number of calls every day. These numbers are mainly dividedinto two categories: advertising harassment and fast-food delivery. One of the most important jobs in telecommunication business is to automatically classify the high-frequency numbers into these two categories using machine learning techniques in the server side.However, the number of advertising harassment is much higher than that of fast-food delivery, leading to a very low recognition accuracy of fast-food delivery numbers. Therefore, this paper employs the Diversified Sensitivity-based Undersampling (DSUS) to handle the class imbalance issue in the classification process and to improve the recognition accuracy of the fast-food delivery numbers. Experimental studies show that, compared to two opponent methods, the DSUS yields a higher recognition accuracy rate of the fast-food delivery numbers.