The College of Computer Science and Information Technology at the University of Basrah organized a scientific lecture entitled:

“Advanced Bayesian Hyperparameter Optimization for XGBoost Using Federated and Fuzzy-Based Frameworks”.

The lecture aimed to highlight the importance of hyperparameter optimization in the XGBoost algorithm, given the direct impact of these parameters on model accuracy, stability, generalizability, and execution time. The lecture also demonstrated that relying on default settings or traditional search methods may not be sufficient for many machine learning problems due to the vastness of the search space and the interconnectedness of the parameters.

The lecture, presented by researcher Nabaa Al-Sunaid, included an explanation of two proposed frameworks for optimizing hyperparameter optimization. The first framework, FHBO, employs an initial federation stage to generate exploratory knowledge from multiple agents, and then uses this knowledge as an enhanced starting point for centralized Bayesian optimization. The second framework, FABO, focused on integrating a fuzzy controller into the Bayesian optimization process to balance exploration and exploitation during the search. The researcher also reviewed the results of applying both frameworks to a set of classification and regression data, comparing them with several well-known optimization methods, and demonstrating their impact on improving predictive performance, reducing execution time, and enhancing the reliability of the results.