Document Type: Original Article
Dept. Of Management, Faculty of Social Science and Economics, Alzahra University, Tehran, Iran
School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran, Iran
Finding an optimized place is undeniably a momentous subject in establishing the marketing strategies of a retail store. Based on the existing literature, the process of selecting an optimized location for a business can be defined as a ranking problem that compares and rates existing or potential sites based on their ability to attract customers. Consequently, this article is concentrated on the evaluation of machine learning ranking methods in ranking existing retail stores based on the data derived from LBSNs. Using feature engineering techniques, we defined and calculated a set of features for 239 retail store branches in Tehran, from the venue data obtained from the Foursquare API. Additionally, we derived a rank for each store representing store popularity via user-generated data from Foursquare, Dunro, and Google Maps. Next, we implemented a number of classification and “learn-to-rank” algorithms to rate these stores. Finally, by evaluating the prediction precision and ranking precision of the algorithms used, we analyzed the fit and prediction power of all ranking algorithms. The outcomes of this research suggest that most algorithms used are, in fact, reliable methods for ranking retail store sites. Therefore, such algorithms can be used as a technique for retail store site selection, given a list of existing or potential sites for a store. Additionally, our results clearly suggest a superiority in the ranking precision of “learn-to-rank” algorithms for retail store placement. Out of all algorithms used, with a ranking precision of 0.854, MART is the most powerful algorithm for ranking retail store sites.