A Spatially Explicit Machine Learning Model for Visit Prediction to Georgia Recreational Vehicle Parks

Keywords: RV Parks, Spatially Explicit Machine Learning, Artificial Neural Network, Multiclass Classification
Abstract Type: Paper Abstract

Abstract

According to the Recreational Vehicle (RV) Industry Association, more U.S. travelers turned to RV travel during the COVID-19 pandemic to avoid human contact in traditional options like hotels. To estimate location-based RV park visitorship across geographic space and time, we use SafeGraph data and present a spatially explicit machine learning (ML) model to predict RV park visits based on visit patterns from large-scale anonymous mobile phone records collected from the RV parks and campgrounds in Georgia, US from January 2018 to April 2022.

We treat the RV park visit prediction as a spatiotemporal multiclass classification problem. We aggregate the origins of visitor flows to the census block group (CBG) level. Then we use the Space2Vec location representation learning model to encode the locations of each CBG and the temporal embedding model to capture the temporal information. The socioeconomic and demographic factors are collected and encoded as a semantic embedding to describe the characteristics of each CBG. We further concatenate the learnable location, temporal, and the CBG’s semantic embeddings to characterize the spatial and temporal features of each trip. A spatially explicit ML model isdeveloped to conduct RV park visit prediction.

Preliminary findings reveal that people in areas with a higher youth population show more tendencies to travel long-distance in summer. In addition, we examine and summarize several measures of the origins such as median household income and minority population, that affect people’s visit decisions before and during the pandemic.