CIS 563 Predicting the outcome of asylum case applications from UNHCR Refugee Data Completed By: ANUSHA PRAKASH SAJIAH NAQIB SANGEETHA VENKATESAN
Introduction There are over 80 million people from every war torn part of the world who have been forced to leave their homes and seek asylum elsewhere. In the US alone, there are 470,000 pending asylum applications and a backlog of over 300,000. Our main goals are to use exploratory data analysis to represent those figures from a dataset of refugees gathered from the years 1999-2017 by the UNHCR to advocate for and provide awareness in the form of exploratory data analysis and data visualization, and use various machine learning models including classification, regression and clustering algorithms to help predict the status of asylum cases and the number of accepted or rejected asylum cases, and compare accuracies from different models.
Prior Work In June 2017, a research paper titled “Can Machine Learning Help Predict the Outcome of Asylum Adjudications?” was published by Daniel L. Chen and Jess Eagle, where the authors analyzed 492,903 asylum hearings in the US, gathered from the Transactional Records Access Clearinghouse. They used a random forest classifier to classify an application as granted or denied, and obtained an accuracy of 79%. Our work is inspired by the paper in 2017, where we aim to identify whether an asylum case is accepted or rejected on a dataset gathered from UNHCR from asylum applications from the years 1999 to 2017. Our paper not only focuses on classification models to classify each record into an accepted asylum case or rejected, but also regression models to predict the number of accepted or rejected applicants. We also aim to compare different classification and regression models and identify the ones best suited for our dataset.