Dr. Vasile Palade (Coventry University, UK)
Dr. Vasile Palade has joined the Department of Computing at Coventry University, United Kingdom, in September 2013, after working for several years with the Department of Computer Science at the University of Oxford. His research interests spans across several machine learning and computational intelligence domains, and include neural networks and neuro-fuzzy systems, different nature inspired learning and optimization algorithms, hybrid intelligent systems, class imbalance learning. Main application areas are bioinformatics and computational biology problems, fault diagnosis, process modelling and control, web usage mining and social network data analysis, image processing. He has published several books and 120 papers in machine learning journals and conference proceedings. He is acting as an Associate Editor to several journals, e.g., Knowledge and Information Systems, Neurocomputing, International Journal of Artificial Intelligence Tools, International Journal of Hybrid Intelligent Systems.
Title Class Imbalance Learning
Abstract
Class imbalance of data is commonly found in many data mining tasks and machine learning applications to real-world problems. When learning from imbalanced data, the performance measure used for model selection plays a vital role. The existing and popular performance measures used in class imbalance learning, such as the Gm and Fm, can still result in sub-optimal classification models. The talk will first present a new performance measure, called the Adjusted Geometric-mean (AGm), which overcomes the problems of the existing performance measures when learning from imbalanced data. Support Vector Machines (SVMs) has become a very popular and effective machine learning technique, but which can still produce sub-optimal models when it comes to imbalanced datasets. The talk will then present FSVM-CIL (Fuzzy SVM for Class Imbalance Learning), an effective method to train FSVMs with imbalanced data in the presence of outliers and noise in the data. Finally, some efficient resampling methods for training SVMs with imbalance data will also be discussed in the context of applications. .
