Please join us for the Statistics Seminar scheduled from 11:30 a.m. to 1 p.m. on Friday, October 18.
Speaker: Professor Mohammad Jafari Jozani, University of Manitoba, Canada
Title: Rank-Based Support Vector Machines for Addressing Imbalanced Binary Classification Challenges
Abstract:
We introduce a novel methodology to tackle the challenge of imbalanced binary classification by leveraging rank information. Our approach employs a maxima nominated sampling (MaxNS) strategy, which biases the training dataset towards the minority class by selecting observations with the highest likelihood of belonging to the minority class from a small, randomly drawn subset of the population. This sampling method, guided by expert opinion, has been largely overlooked in the machine learning literature. To effectively integrate the rank information from MaxNS into the learning process, we propose new rank-based Hinge and Logistic loss functions, specifically designed to capture the added rank information from MaxNS samples. Building on these, we develop MaxNS-based Support Vector Machines (SVMs) and introduce efficient algorithms to solve the resulting learning problems. Extensive numerical studies are conducted to demonstrate the performance and effectiveness of the proposed methods.
Zoom information:
Meeting ID: 913 4599 2612
Passcode: 40815977
