Predicting Animal Shelter Pet Adoption Times and Feature Importance Analysis using CatBoost

Cem Sazara, Xin Gao

Abstract: Animal shelters in the US provide care for stray or unwanted animals. Animal shelters have become crucial for the welfare of the animals brought to these facilities. This study uses the intake and outcome data provided by the Austin Animal Center from 2013 to July 2022. Main focus of the paper is to study different factors to determine possible stay times of animals in the shelter until adoption and build a predictive model for the future adoption time. For this purpose, first, the intake and outcome data are put into single table format. Then, adoption times are divided into two major groups: Adopted in less than 15 days and more than 15 days. To measure the impact of different attributes, a tree-based boosting method is used. This method is trained, validated and tested with the corresponding splits of the dataset. At the end, we provide an in-depth analysis of the attributes leading to different stay times. We also provide a GitHub link <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">‡</sup> for the dataset and the trained models in this paper.