Regency D
Site Selection and Statistical Learning (303972)
Sima Sharghi, Bowling Green State University
*Kevin Edward Stoll, Bowling Green State University
Keywords: site selection, statistical learning, propensity score, predictive modeling, binning, density estimation
Based on current investments, we use binning and statistical learning to define and indicate new markets for real estate-heavy businesses and location-based services. We creatively define markets based on demographic and competition estimates of disjoint regions surrounding a site and outlined by concentric drive-time polygons. This creates explanatory variables such as `the estimated number of households with household income between \$75,000 and \$85,000 within 0 to 10 minutes of driving' and ‘the number of competition sites within 20 to 30 minutes of driving' enabling statistical learning to discover the preferred markets based on important similarities with current markets. As a case study, we explore the intrinsic relationships between surrounding demographics and competition for three major membership-only warehouse clubs. We apply statistical procedures of binning, dimension reduction, density estimation, predictive modeling, and propensity scoring. For each technique and warehouse chain, we create and share heat-maps of the top 50 most populated core-based statistical areas which depict the locations that are most and least likely to attract the specified warehouse chain.