Research on Developing a Telecom Fraud Detection Model Based on Logistic Regression

This research addresses the growing threat of telecom fraud amid financial industry expansion. Traditional detection methods struggle with evolving fraud tactics, necessitating AI and machine learning innovations. Utilizing Baltamatica, an innovative generic scientific computing software, this research explores its algorithmic capabilities, conducts professional model development practice and employs Logistic Regression to develop an efficient telecom fraud scoring model. The model achieves a KS of 58%, 87.57% accuracy, and a 51% recall rate. After score transformation, the model effectively identifies and intercepts high-risk customers, demonstrating strong predictive power. Additionally, the study compares Baltamatica's modeling process with industry standards and recommends enhancements such as VIF, WOE, and XGBoost integration. These findings support fraud prevention strategies in financial institutions while aiding the advancement of innovative scientific computing software.