Title: (Almost) real time outlier detection via principal least squares support vector machines Authors:  Andreas Artemiou - University of Limassol (Cyprus) [presenting]
Abstract: The aim is to propose an influence measure for outlier detection using principal least squares support vector machines (PLSSVM). A number of papers have discussed the influence measure of sufficient dimension reduction (SDR) methodology, but they focus on inverse moment-based (SDR) methods. The influence measure for an SVM-based SDR method is developed for the first time. Also, given that the PLSSVM algorithm was originally proposed in the online dimension reduction framework, it is demonstrated that this influence measure can be applied in an online outlier detection framework.