Authors: Chanwoo Jin*, San Diego State University / UC Santa Barbara
Topics: Geographic Information Science and Systems, Spatial Analysis & Modeling
Keywords: GeoAI, RNN, survival analysis, restaurants, South Korea
Session Type: Paper
Presentation File: No File Uploaded
The opening and closing of restaurants are significant signs of change in a city. Identifying underlying spatiotemporal patterns deepens our understanding of urban dynamics, but also enables planning and public policy to play a role in fostering economic growth and development. Although many studies have attempted to explain factors that contribute to survival of restaurants and predict their lifespans, traditional methods show less accurate prediction because it is difficult to determine the relationships between socioeconomic factors and their survivability with limited obtainability of data and human knowledge. To address this issue, this study proposes a deep learning model, recurrent neural network (RNN), with a simple variable set to improve prediction of restaurant survivability. Deep learning has been applied to various scientific disciplines and has shown accurate predictions. Geographers have also recently used deep learning techniques such as a convolutional neural network (CNN) to classify geographic objects from remotely sensed images or places from text. Nevertheless, compared to CNN, RNN has been less applied to spatial events even though it is significantly useful for sequential data and highly non-linear relationships between past and present. It is, therefore, useful to apply RNN to survival analysis on restaurants to understand spatiotemporal complexity of their entrepreneurship. For an empirical study, we analyze open public data on all approval businesses in South Korea that includes coordinates, current status, starting and closing date. Even only with relative distances between restaurants, RNN can predict survivability of restaurants more accurately than a traditional survival analysis, cox regression.