A Random Forest Regression Model for Predicting the Movement of Horseshoe Crabs in Long Island Sound

Abstract

Developing models to predict animal movement patterns is an important area of study in ecology and wildlife management. Project Limulus, a community research program, has been tracking the movement of tagged American horseshoe crabs (Limulus polyphemus) in Long Island Sound since 1998. During the spawning season, horseshoe crabs are captured by hand, tagged and then released. Recaptured horseshoe crabs give valuable information about their behavior and movement patterns. In this paper, we tested various models to find the best predictor for the movement of horseshoe crabs based on the observed movement activity in previous years. We experimented with three different models: Linear Regression, Decision Tree, and Random Forest Regression models. We used the data for 2018 as our test set and the data of all previous 19 years as our training set (19,219 records). The Random Forest Regression model proved to be the best predictive model and resulted in the smallest RMSE and MAE, as well as the smallest maximum error in prediction. The predicted horseshoe crab locations can be targeted in the next season for recapturing tagged horseshoe crabs. It also concentrates the scientists’ effort and time to find the maximum number of horseshoe crabs.