.png){kind=icon}
.png){kind=icon}
Return
CGTCTR: telecom fraud detection based on data imputation and integrated deep optimization classification model
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.130668.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Telecommunication fraud poses a serious threat to social security due to its cross-regional and intelligent characteristics. However, the actual telecommunication data often have problems such as missing values, high-dimensional nonlinearity and spatio-temporal dynamic correlation, resulting in limited accuracy of traditional detection methods. In view of this, this study proposes a telecom fraud detection model that combines data imputation and integrated deep classification to improve the detection accuracy and robustness in complex scenarios. Firstly, an channel-spatial attention mechanism combined generative adversarial interpolation network (CSAM-GAIN) model is designed for data imputation. The spatio-temporal correlation of data is captured by series channel and spatial attention mechanism, and the completion data close to the real distribution is generated by wighted loss optimization, so as to solve the problem of insufficient capture of missing value correlation in existing methods. Secondly, an integrated classification model is constructed. The Convolutional Time Convolutional Network module extracts local spatial features and temporal features,respectively. The channel-spatial attention inception mechanism module enhances feature discrimination through multi-scale residual structure and mixed attention. The SwinTransformer module captures global spatio-temporal dependencies and makes up for the limitations of local feature extraction. Experiments verify the performance of the model. The results show that our proposed model is superior to baseline methods in terms of accuracy,the integrated classification model improves the accuracy by 2%-5% compared with the mainstream models. This study provides a solution that takes into account both data integrity and feature depth mining for telecom fraud detection, which is of great significance to promote the practical application of intelligent anti-fraud technology.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W