A novel framework for facility location evaluation

Is a given location appropriate to set a facility, and if it is, how will this facility perform? This is denoted as the facility location problem, which has attracted interests of academic researchers and industrial practitioners. This problem is traditionally considered in spatial analysis and forecasting literature. However, existing methods can hardly handle the problem well because of the fragmented information and insufficient training data. In this paper, we present a novel framework which combines spatial analysis and forecasting analysis for facility location. Firstly, a classifier is built on spatial information to evaluate locations' environmental patterns. For each pattern, a predictor is then constructed to predict facility performance. Besides, we also propose a data aggregation method to pre-process raw spatial data, and a semi-supervised learning method to expand insufficient training data. Experimental results of a case study demonstrate the effectiveness of the framework on supporting real-world facility location decisions.