Evaluating XGBoost for balanced and Imbalanced datasets
The talk will introduce XGBoost, and provide examples of evaluation metrics for ML models in fraud and risk-scoring applications.
Globally, fraud accounts for more than 2 percent of the total revenue loss in the telecommunication sector. Fraud detection is not a trivial problem: fraudsters try different tactics to simulate customers' behavior, constantly invent new ways to fool the system, and might appear sporadically. The project could help prevent equipment theft, reselling, and commission fraud among the many types of fraud scenarios. Dr. Gissel Velarde will present an AI-based Fraud Detection System, including several experiments with XGBoost, demonstrating how efficient machine learning models can automatically detect fraud cases depending on the data used.
Gissel Velarde
Dr. Gissel Velarde holds a Ph.D. in Computer Science and Engineering from Aalborg University for her work on machine learning-based models to analyze perceptual applications. Dr. Velarde has received several prizes and awards as an engineer and data scientist. In 2021, she received a teaching award at UPB for her lecture: Selected Topics in Artificial Intelligence. In 2022 and 2023, she received Vodafone Start Awards for contributing to AI Strategies and the execution of experiments that generated knowledge to develop a robust detection system. Currently, she is a Senior Expert Data Scientist at Vodafone, leading the risk assessment project from the AI and machine learning side.
