Forecasting venue popularity on location-based services using interpretable machine learning

Customers are increasingly utilizing location-based services via mobile devices to engage with retail establishments. The focus of this paper is to identify factors that help to drive venue popularity revealed by location-based services, which then better facilitate companies’ operational decisions, such as procurement and staff scheduling. Using data collected from Foursquare and Yelp, we build, evaluate, and compare a wide variety of machine learning methods including deep learning models with varying characteristics and degrees of sophistication. First, we find that support vector regression is the best performing model compared to other complex predictive algorithms. Second, we apply SHAP (Shapley Additive exPlanations) to quantify the contribution from each business feature at both the global and local levels. The global interpretability results show that customer loyalty, the agglomeration effect, and the word-of-mouth effect are the top three drivers of venue popularity. Furthermore, the local interpretability analysis reveals that the contributions of business features vary, both quantitatively and directionally. Our findings are robust with respect to different popularity measures, training and testing periods, and prediction horizons. These findings extend our knowledge of location-based services by demonstrating their potential to play a prominent role in attracting consumer engagement and boosting venue popularity. Managers can make better operational decisions such as procurement and staff scheduling based on these more accurate venue popularity prediction methods. Furthermore, this study also highlights the importance of model interpretability which enhances the ability of managers to more effectively utilize machine learning models for effective decision-making.