2015 Informs Annual Meeting

TB30

INFORMS Philadelphia – 2015

TB31 31-Room 408, Marriott Connected Vehicle Analytics

3 - An Approach to Estimating Customer Lifetime Values for Apartment Tenants Jian Wang, Vice President, Research & Development, The Rainmaker Group, 4550 North Point Parkway, Alpharetta, GA, 30022, United States of America, jwang@letitrain.com Estimating tenant lifetime values is important for apartment revenue management. We propose a heuristic approach to predicting renewal likelihoods and estimating tenant lifetime values. We then present empirical results based on real apartment data. TB30 30-Room 407, Marriott Intelligent Agents and Systems Contributed Session Chair: Mohsen Moghaddam, PhD Candidate, Purdue University, 1155 Anthrop Dr., Apt. 9, West Lafayette, IN, 47906, United States of America, mmoghadd@purdue.edu 1 - Optimizing Physician to Patient Consults using Robot-based Virtual Systems Henry Ibekwe, Post Doctoral Researcher, Independent, Richmond, United States of America, hibekwe1@gmail.com The delivery of quality healthcare for chronically ill patients is burdened by the limited resources available to physicians and healthcare facilities. We propose the use of virtual-presence autonomous robot systems to optimize the physician to patient consultation by minimizing the patient wait-time and maximizing the number of physician consults given limited resources. We formulate a robot- patient interaction model as a stochastic process and solve using discrete-time dynamic programming. 2 - A Study on the Influence of Trust and Distrust Ratings in Social Networks on Cold Start Users Sanjog Ray, Assistant Professor, Indian Institute of Management Indore, Rau Pithampur Road, Indore, 453331, India, sanjogr@iimidr.ac.in This study examines how cold start users get influenced by the trust and distrust scores of other users in a social network. We examine the users trusted by cold start users on the basis of critical parameters: number of trust statements, number of distrust statements, and number of items rated. We base our findings on our analysis of the real life Epinions dataset. Our analysis has implications for design of trust aware recommender systems for cold start users. 3 - Analyzing Inventory Policies in Multi-stage Automatic Manufacturing Systems Barin Nag, Professor, Towson University, Department of E-Business & Technology Ma, Towson, MD, 21252, United States of America, bnag@towson.edu, Dong-qing Yao, Sungchul Hong In a multi-stage manufacturing system each stage fills demand from any combination of buffer inventory or production. Lowest inventory levels may not be lowest cost, with contradictions arising from the costs of delays of physical production, backlogs, breakdowns, and bottlenecks. We study best performance inventory policies using varied production architectures. 4 - A Modeling Framework of Cyber-Physical Systems Ashutosh Nayak, Student, Purdue University, 318 N Salisbury St, Apt. 8, West Lafayette, IN, 47906, United States of America, nayak2@purdue.edu, Shimon Y. Nof, Seokcheon Lee, Rodrigo Levalle Effective modelling of CPS is a big challenge. In this work, we propose a resource sharing based framework for CPS aimed at maximizing its utility. This framework represents CPS as a network of tasks and resources characterized by utility functions and overlapping resource communities. A distributed control approach backed by utility aggregation function is considered for optimality and stability. Its implementation is illustrated through two examples: Smart factory and multi- robot system. 5 - Collaborative Networked V-organizations: Design & Integration Mohsen Moghaddam, PhD Candidate, Purdue University, 1155 Anthrop Dr., Apt. 9, West Lafayette, IN, 47906, United States of America, mmoghadd@purdue.edu, Shimon Y. Nof Modern distributed, networked, and collaborative organizations of humans/machines/firms enable systematic integration of distributed resources for processing dynamic/diverse tasks. We design collaborative networked v- Organizations by integrating physical (location of resources) and virtual (allocation of tasks) dimensions, for higher service level, stability, and utilization. A mixed-integer program and a tabu search are developed for modeling and optimization purposes, respectively.

Sponsor: Data Mining Sponsored Session Chair: Juan Li, Member of Research Staff, Xerox Innovation Group, 800 Phillips Road, 128-27E, Webster, NY, 14580, United States of America, Juan.Li@xerox.com 1 - A System for Estimating Traffic Congestion Measures in a Network using GPS Smartphone Charles Chung, Vp Products, Brisk Synergies, 295 Hagey Blvd, 1st Flr, Waterloo, Canada, charles.chung@brisksynergies.com A smartphone app is developed for logging route data. A platform is then built for mapping traffic congestion using speed indicators average speed and speed differential at the link level. The results demonstrate the feasibility and huge potential our data collection system that can be implemented in any city and sets the growth for real-time applications for connected vehicles. 2 - Online Travel Mode Identification with Smartphones Qing He, Assistant Professor, SUNY Buffalo, 313 Bell Hall, Buffalo, NY, 14051, United States of America, qinghe@buffalo.edu, Xing Su, Hernan Caceres, Hanghang Tong We propose an online classification algorithm to detect user’s travel mode using mobile phone sensors. Our application is built on the latest Android smartphone with multimodality sensors. By applying a hierarchical classification method, we achieve high accuracy in a binary classification wheelers/non-wheelers travel mode, and all six travel modes. 3 - Locating Heterogeneous Traffic Sensors to Improve Network Surveillance Benefit Optimal placement of traffic sensors is significant to improve urban mobility. In this study, a mathematical optimization model of deploying heterogeneous sensors (i.e. Bluetooth sensor and loop detector) to large-scale traffic network is proposed. Maximizing real-time information report reliability and coverage are chosen as dual objectives. In addition, the effect of real-time GPS-based probe vehicle data is also considered. A case study in Washington D.C. area is conducted for demonstration. 4 - Inferring Trajectories for Partial Observations Juan Li, Member of Research Staff, Xerox Innovation Group, 800 Phillips Road, 128-27E, Webster, NY, 14580, United States of America, Juan.Li@xerox.com, Moshe Lichman, Padhraic Smyth The amount of spatial trajectory data is growing fast with the rapid increased availability of GPS-embedded vehicles. The trajectory data is mixed with high and low sampling rate with partial observations. In this study, we aim to build probabilistic models to infer possible traversed route for low sampling rate vehicle trajectory data. TB32 32-Room 409, Marriott Business Analytics in Higher Education Industry Sponsor: Analytics Sponsored Session Chair: Roger Gung, Director, Business Analytics & Operations Research, University of Phoenix, 3137 E Elwood St, Phoenix, AZ, 85034, United States of America, roger.gung@phoenix.edu 1 - Marketing Mix Optimization Roger Gung, Director, Business Analytics & Operations Research, University of Phoenix, 3137 E Elwood St, Phoenix, AZ, 85034, United States of America, roger.gung@phoenix.edu Marketing spend allocation drives the volume of new marketing inquiries (NMI) and enrollments. Two-stage non-linear regression models were built to formulate NMI channels with respect to marketing spends which were defined as either endogenous, exogenous or instrument variables. The optimization model was formed by aggregating all NMI channels’ regression models into one objective function. The optimal spend allocation was then derived from the model every quarter to guide marketing strategies. Xuechi Zhang, Graduate Research Assistant, University of Maryland, 0147C Eng Lab Bld, University of Maryland, College Park, MD, 20742, United States of America, zhangxc90@gmail.com, Ali Haghani

294