Deer Capture Prediction

A Comparison of Maxent and Machine Learning with Presence-only Data

Abstract

Mammals often cause serious damage to ecosystems, agriculture, forestry, and residential environments. Hunters’ contribution to capturing such harmful mammals is expected. The useful information for hunters will be of great importance for enhancing their effective contribution. However, the prediction of capture sites based on past capture information is scarce in Japan. In this study, we utilized the deer capture records in Shimane Prefecture from 2010 to 2018 by applying high-resolution geodata and machine learning techniques to predict capture probabilities. The results showed that using CatBoost at a 250m mesh significantly improved the prediction performance compared to the ordinary method used in the species distribution prediction in Japan (Maxent at 1000m mesh).