Informs Annual Meeting 2017

SD65

INFORMS Houston – 2017

SD65

SD66

370F Data Mining in Churn Decision Analytics Sponsored: Data Mining Sponsored Session Chair: Roy Jafari Marandi, Mississippi State University, rj746@msstate.edu 1 - A New Algorithm for Segmented Modeling: An Application in Customer Churn Prediction Arno De Caigny, IÉSEG School of Management, France, a.de-caigny@ieseg.fr, Kristof Coussement, Koen W. De Bock Heterogeneity between customers poses a challenge for churn prediction models. Therefore a new algorithm, the logit leaf model, has been developed to better classify data that is characterized by heterogeneous groups. Logit leaf modeling consists of two stages: a segmentation phase and a prediction phase. In the first stage customer segments are identified using decision rules while in the second stage a model is created for every leaf of this tree. This approach is benchmarked against ensembles, statistical approaches and decision trees with regards to the predictive performance and comprehensibility. 2 - Churn Prediction in Telecommunication: Sampling, Feature Selection and Ensemble Methods Adnan Idris, The University of Poonch, Rawalakot, District Poonch, Pakistan, adnanidris@upr.edu.pk Efficient churn prediction is helpful for Telecommunication industry to retain customers. Development of any efficient churn prediction system depends upon the appropriate selection of sampling, feature selection and classification methods due to enormous nature of telecom datasets. In this bid, various feature selection, sampling and classification methods have been explored to efficiently tackle this challenging research problem. Minimum Redundancy and Maximum Relevancy based feature extraction, PSO based intelligent undersampling along with tree based ensembles ,Genetic Programming(GP) and AdaBoosting based ensemble methods attain improved prediction performance. 3 - Profit Driven Comparison of Social Network Analytics Methods for Predicting Customer Churn in TELCO María Óskarsdóttir, KU. Leuven, Naamsestraat 69, Leuven, Belgium, maria.oskarsdottir@kuleuven.be Social network analysis is being used in telco to build profit driven applications for predicting customer churn. In this benchmarking study, a unique number of call-detail record data sets from across the world is used to statistically evaluate the effect of relational classifiers and collective inference methods on the predictive power of relational learners. In addition, using the Expected Maximum Profit measure, we evaluate the performance of various models where relational learners are combined with traditional methods of predicting customer churn in order to find the optimal model. 4 - Customer Decision Making Dynamics in a Multichannel Context Wenyu Jiao, ESSEC Business School, Cergy, France, wenyu.jiao@essec.edu, Nicolas Glady This paper aims at understanding the purchase channel dynamic on the customer-firm relationship and the Customer Lifetime Value(CLV). To this purpose, we address two important research questions in a multichannel context: (i) how does the customers’ channel usage history dynamically impact the components of the CLV, and (ii) are the multichannel customers who start from offline channels and those who start from online channels equally valuable? We propose a hidden Markov model to understand the customer-firm relationship dynamic in a multichannel setting and to estimate the CLV. Our dynamic model is necessary in order to account for the evolution of customer behavior across different states. 5 - Optimum Churn Decision Making by Cost-sensitive SOED ANN in Telecommunication Industry Roy Jafari Marandi, Mississippi State University, 260 McCain Engineering Buildin, Mississippi, Mississippi State, MS, 39762, United States, rj746@msstate.edu, Brian K. Smi Churn prediction can only be as good as the hidden pattern in the dataset used for prediction. Attempting to captures the individuality of each customer before arriving at a prediction decision can bring about another dimension in the improvement of churn prediction decisions. The existing data analysis techniques only extract different patterns existing in datasets and apply them to all of the rows of the data in the same manner. In reality, every row is very different and have a different level of importance when it comes to making a final decision. This presentation will introduce Cost-Sensitive SOED Artificial Neural Network which has assuaged these shortcomings.

371A Facility Layout Sponsored: TSL, Facility Logistics Sponsored Session Chair: Sadan Kulturel-Konak, Penn State Berks, Reading, PA, 19610, United States, sadan@psu.edu 1 - Are Curved Racks Better than Straight Racks for Retail Layouts? Pratik J. Parikh, Wright State University, 3640 Colonel Glenn Highway, 207 Russ Engineering Center, Dayton, OH, 45435-0001, United States, pratik.parikh@wright.edu, Bradley R. Guthrie Because retail shopper’s will only purchase products on a shopping trip that they see, designing retail layouts with product visibility as a key priority has been widely discussed in literature. To this extent, we have developed analytical and computational models for the dynamic interaction between a shopper’s field of view and a layout of racks in 3D. We focus on two design variables: (i) rack orientation and (ii) rack curvature. Based on an experimental study with varying values of shopper head rotation, rack height, and traffic patterns, we find that curved racks can offer an improvement in exposure over traditional straight racks. 2 - Placement of M Finite-size Rectangular Facilities in an Existing Layout: Lower Bounds and Dominance Results Rakesh Nagi, University of Illinois at Urbana-Champaign, Department of Industrial & Enterprise Systems, 117 Transportation We study the problem of optimally placing M new finite size rectangular facilities (NFs) in a layout with N existing rectangular facilities (EFs). Interactions are present between different pairs of EFs and NFs, serviced through the Input/output points located on the facility boundary. We present a dominance procedure utilizing different lower bounding techniques, which is shown to outperform the explicit enumeration procedure. 3 - Dynamic Facility Layout with Input / Output Points Sadan Kulturel-Konak, Penn State Berks, Tulpehocken Road, P.O. Box 7009, Reading, PA, 19610, United States, sadan@psu.edu When the locations of the departments in a facility have to be decided where there are multiple planning periods and the interdepartmental material flows change over the planning periods, the problem in concern is called the Dynamic Facility Layout Problem (DFLP). The main challenge in the DFLP is that there are two conflicting objectives of the problem: minimizing the material handling cost and the rearrangement cost. The proposed approach, a new matheuristic, considers determining relative department locations, their dimensions, as well as input/output (I/O) points concurrently for the first time in the literature. 4 - Optimal Unit Load Warehouse Designs by using Visibility Graph Building, MC-238, Urbana, IL, 61801, United States, nagi@illinois.edu, Aishwarya Anandan, Ketan Date

Sabahattin Gokhan Ozden, Pennsylvania State University, Jenkintown, PA, 19046, United States, sgo7@psu.edu, Alice E. Smith, Kevin R. Gue

We introduce the visibility graph as a new way of estimating the length of a route traveled by order pickers in a warehouse. Following aisle centers leads to longer travel distances when an order picker picks items within picking aisles which have angles other than 90 degrees between cross aisles. We present optimal/near- optimal designs by solving single and double cross aisle designs. We compare them with the optimal designs by using aisle centers as the length of a route estimation method (i.e. Chevron and Leaf layouts).

SD67

371B Panel Discussion on Publishing in QSR Journals: The Editors’ Perspective Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Adel Alaeddini, University of Texas at San Antonio, San Antonio, TX, 78249, United States, adel.alaeddini@utsa.edu 1 - Publishing in QSR Journals: the Editors’ Perspective Kamran Paynabar, Georgia Institute of Technology, Georgia Tech, H. Milton Stewart School Of Isye, Atlanta, GA, 30332-0205, United States, kamip@umich.edu This panel brings QSR journal editors to provide their perspectives and experiences with the audience, and answer their questions pertaining to publication in these journals.

124