Enhancing facility locating via a novel hybrid model

Given its importance, facility location has long attracted many research efforts from both the academic and industrial areas. However, due to the cost and privacy issues, researchers usually suffer from the lack of business information which is crucial for the decision makings. Recent years have witnessed the explosion of geographical information systems (GISs), and the spatial information provided by GISs becomes a valuable supplement to the limited business information for facility location decision. Along this line, in this paper, we present a hybrid model which combines spatial analysis and forecasting analysis to solve this problem. That is, a classifier is built first on the spatial data to evaluate environmental conditions of the location. Then based on the classification results, a predictor is built on both the spatial and business data to predict the facility´s performance. To deal with the problem of missing much business information while building the classifier, we also propose a semi-supervised learning method to expand the training data set. Finally, experimental results on a case study demonstrate that the hybrid model indeed shows merits on supporting real-world facility location decisions.