2016 INFORMS Annual Meeting Program

WD54

INFORMS Nashville – 2016

3 - Bank Branch Operational Performance Through a Robust Multivariate And Fuzzy Clustering Approach Oscar Albeiro Herrera-Restrepo, PhD. Industrial and Systems Engineering, Virginia Tech, 4339 Taney Avenue, Apt 401, Falls Church, VA, 22304, United States, oscar84@vt.edu, Konstantinos P Triantis, William L. Seaver We propose a multi-step procedure that integrates fuzzy clustering analysis and data envelopment analysis (DEA) to group bank branches into managerial clusters and to investigate their operational performance. We build and expand on previous research by including fuzzy clustering. We look for changes in clustering composition due to branches belonging to multiple clusters, and changes in operational efficiency performance due to fuzzy clustering. All this while looking at influential branches. Our premise is that fuzzy clustering allows differentiating clusters beyond scale/size, and that it affects operational efficiency performance. WD54 Music Row 2- Omni Smart Services: Design, Development, and Measurement Sponsored: Service Science Sponsored Session Chair: Robin Qiu, Penn State, 30 E. Swedesford Road, Malvern, PA, 19355, United States, robinqiu@psu.edu Co-Chair: Hi-Hun Kim, Pohang University of Science and Technology, Pohang, Korea, Democratic People’s Republic of, kh_kim@pohang.ac.kr 1 - Identifying New Service Opportunities For Driving Safety Enhancement Based On Driving Behavior Analysis: Commercial Vehicle Case In Korea Chang-Ho Lee, Pohang University of Science and Technology, 77 Cheongam-Ro. Nam-Gu., Pohang, Korea, Republic of, dlckdgh@postech.ac.kr, Min Jun Kim, Young-Mok Bae, Kwang-Jae Kim The goal of this research is to identify service opportunities for enhancing driving safety for commercial vehicles (including intra-city buses, express buses, and trucks). Based on an analysis of vehicle operational data in conjunction with accident data, new service opportunities for enhancing driving safety are identified. The service opportunities would contribute to developing new services for commercial vehicle companies and related authorities in Korea. 2 - Development Of A Daily Health Behavior Index For College Students Ki-Hun Kim, Pohang University of Science and Technology, Pohang, Korea, Republic of, kh_kim@postech.ac.kr, Kwang-Jae Kim Recently, a smart wellness service has been developed to support daily wellness management for college students. During a day, the service collects health behavior (activities, sleep, and diet) data of a student via smart devices. As part of the service, a daily health behavior index was developed to evaluate the student’s health behaviors based on the collected data. Daily health behavior data of 47 college students were collected during a four-week experiment and used to develop the index. This talk presents how the index was developed and utilized in the service.

2 - The Effect Of Direct Marketing On Online Purchases – An Empirical Study Xingyue Zhang, Lehigh University, 621 Taylor Street, Bethlehem, PA, 18015, United States, xiz313@lehigh.edu, Yuliang Yao Use a unique dataset collected from one of the largest classified ads website in China, we empirically examine the effect of offline call intensity on online customer purchase probability and the carryover effect of call intensity. We find that online customer purchase probability is increasing in call intensity but at a decreasing rate. In addition, there exists a strong carryover effect where the call intensity in the past 4 weeks does not fade away but have a positive effect on recent customer purchase. Our estimations show that both too much or too little call intensity will result in considerably worse outcomes. 3 - Product Recommendations In E-mail Marketing: A Randomized Field Experiment Ting Li, Erasmus University, T09-14, Burg Oudlaan 50, Rotterdam, 3000DR, Netherlands, tli@rsm.nl, Dimitrios Tsekouras Websites retarget their customers by sending e-mail with personalized product recommendations based on their past browsing behavior and preferences. In this study, we examine the effectiveness of such e-mail communication across two types of product recommendations: content-based recommendations and visual recommendations. We conducted a large-scale randomized field experiment to investigate how these effects vary depending on customers’ purchase stage and the temporal distance between their last website visit and receiving the email. The study provides insights for e-commerce websites regarding the improvement of their e-mail retargeting campaigns. Chair: Koki Ho, University of Illinois at Urbana-Champaign, 302F, Talbot Laboratory, 104 S. Wright St, Urbana, IL, 61801, United States, kokiho@illinois.edu 1 - A Decision Framework For Uber-like Transportation Platform Peiyu Luo, PhD Student, University of Louisville, Louisville, KY, 40217, United States, p0luo002@louisville.edu, Lihui Bai The rising platforms such as Uber, Lyft and Sidecar empower individuals to provide short-range point-to-point ridesharing services other than traditional transportation services (e.g. bus, subway, taxi). This paper aims to provide decision support tools for such peer-to-peer transportation businesses. We divide the business operations into three stages: resource identification, resource allocation and task assignment, and use optimization models and prospect theory in decision making to formulate the three stages. Computational results will be reported. 2 - Leveraging Machine Learning To Support Agricultural Decision-making Emily Burchfield, PhD Candidate, Vanderbilt University, Nashville, TN, 37212, United States, emily.k.burchfield@vanderbilt.edu, John J Nay, Jonathan Gilligan This project applies machine learning to remotely sensed imagery to train and validate predictive models of vegetation health. We processed eleven years of NASA MODIS data and applied gradient boosted machines to the lagged data to forecast future values of vegetation health. We assessed the predictive power of our model across space, time, and land use categories. Our models have significantly more predictive power on held-out datasets than simpler baselines. We constructed an open source tool that predicts per-pixel vegetation health in a user-specified region of interest at 16-day intervals. This tool is useful in regions where clouds prevent real-time monitoring of vegetation dynamics. 3 - Profit Optimization For Rare Earth Permanent Magnet Value Recovery Under Supply And Demand Uncertainties Hongyue Jin, PhD Candidate, Purdue University, 2101 Cumberland Ave Apt 2107, West Lafayette, IN, 47906, United States, jin156@purdue.edu, Yuehwern Yih, John Sutherland Rare earth permanent magnets (REPMs) play an essential role in green energy production and yet face a significant supply risk. To alleviate the risk, value recovery from end-of-life product is proposed. This research develops an inventory management strategy for REPM recovery under market supply and demand uncertainties. A linear programming (LP) model is then developed to find an upper bound for the proposed strategy. Several scenarios are evaluated with a hard disk drive (HDD) example. The proposed strategy helps increase the profit, and the performance is well comparable to the upper bound. WD56 Music Row 4- Omni Decision Support Systems II Contributed Session

WD55 Music Row 3- Omni E Business/Commerce I Contributed Session

1 - On-site Personalized Product Recommendations: A Field Study Dimitrios Tsekouras, Erasmus University, Burgemeester Oudlaan 50, 3062PA, Rotterdam, P.O. Box 1738, Netherlands, dtsekouras@rsm.nl, Ting Li Consumers receive on-site personalized product recommendations about alternative products based on their past behavior. In this paper, we study the effectiveness (click and purchase) of these recommendations based on products (1) browsed, (2) put on wishlist, or (3) bought, using a large dataset from a major e-retailer. We examine the extent to which these effects differ for products with different price levels, in different product categories, as well as depending on how long after the initial interaction with the product source are the recommendations presented. The findings provide suggestions for improving the on-site recommendation for e-commerce websites.

472