Modeling feral swine distribution in Georgia using logistic and autologistic regression

Abstract

Feral swine (Sus scrofa) is a very destructive exotic mammal in the United States that carries pathogens of several diseases, endangers the safety of human beings, and disturbs local ecosystems. Although efforts have been made to monitor the distribution of feral swine, little has been done to model and predict the future distribution of feral swine. This project aims at tackling this problem by identifying the relationship between feral swine distribution and a series of environmental and cultural factors based on logistic regression. To assess the effects of spatial autocorrelation in modeling feral swine, autologistic regression was also applied to be compared with the ordinary logistic regression. The results suggest the autologistic regression model is superior to the ordinary counterpart with better performance. In addition, it is strongly recommended that the ordinary logistic regression methods should be employed with caution when spatial autocorrelation exists because they may yield misleading results.