A Mixed-integer Programming Approach to Example-dependent Cost-sensitive Learning
Mustafa Gökçe Baydoğan, Boğaziçi University
This research studies example-dependent cost-sensitive learning that brings about varying costs/returns based on the labeling decisions in classification tasks. Originating from decision-making models, these problems are distinguished in areas where cost/return information in data is focal, instead of the true labels. For example, in churn prediction and credit scoring, the primary aim is to build predictive models that minimize the misclassification error. Then, the outputs of the model are used to make decisions to minimize/maximize the costs/returns. In other words, prediction and decision making are considered as two separate tasks which may provide local optimal solutions. To resolve such problems, we propose a general strategy to incorporate instance-based costs/returns in a learning algorithm. Specifically, the learning problem is formulated as a mixed-integer program to maximize the total return. Given the high computational requirements of the mixed-integer linear programming problems, this model can be practically inefficient for training on large-scale data sets. To address this, we also propose Cost-sensitive Logistic Regression, a nonlinear approximation of the formulated linear model, which is efficiently solvable. Our experimental results show that the proposed approaches provide better total returns compared to traditional learning approaches. Moreover, we show that the optimization performance of the mixed-integer programming solver can be enhanced by providing initial solutions from Cost-sensitive Logistic Regression to the mixed-integer programming model. This is a joint work with Tarkan Temizöz.
Mustafa Gökçe Baydoğan is an assistant professor in Department of Industrial Engineering at Boğaziçi University, Istanbul, Turkey. Before joining Boğaziçi University, he worked as a postdoctoral research assistant in the Security and Defense Systems Initiative at Arizona State University (ASU) between 2012-2013. He received his Ph.D. degree in Industrial Engineering from ASU in 2012. His B.S. and M.S. degrees are in Industrial Engineering both from Department of Industrial Engineering at Middle East Technical University, Ankara, Turkey in 2006 and 2008 respectively. His research interests focus on statistical learning, with applications in spatio-temporal data mining and data mining for massive, multivariate data sets. Details about him and his work can be reached through: www.mustafabaydogan.com.
Friday, June 10, 2022, 4.00 pm - Zoom Meeting