American graduate admissions: both sides of the table

Gupta, Narender

Permalink

https://hdl.handle.net/2142/92866 Copy

Description

Title

American graduate admissions: both sides of the table

Author(s)

Gupta, Narender

Issue Date

2016-07-18

Director of Research (if dissertation) or Advisor (if thesis)

Roth, Dan

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Graduate Admissions
• Machine Learning
• Latent Variable

Abstract

This is a comprehensive study of graduate admission process in American universities. There are multiple entities involved in the process, out of which the most significant ones are: • The candidate applying for admission in a department in a school • The decision-makers acting upon the candidates' application The goal of this study is to understand the admission process from each of these entities' perspective, and provide them decision-support models for their respective tasks. Although both of the entities interact through a common set of datapoints, i.e. candidate admission application, each of them works towards a very different goal. The juxtaposition of these two tasks provides a very interesting challenge which is hard to resolve deterministically. Solution to such a problem requires learning techniques which can find patterns, adapt according to the dynamic nature of problem, and produce results in a probabilistic fashion. We study and model the graduate admission process from a machine learning perspective based on analysis of large amounts of data. The analysis considers factors such as standardized test scores, and GPA, as well as world knowledge such as university similarity, reputation, and constraints. Based on the targeted entity, learning problem is formulated as classification problem or ranking problem. During learning and inference, not only those features are considered which are available from the data directly, but also the hidden features which need to be incorporated generatively. Our experimental study reveals some key factors in the decision process and, consequently, allows us to propose a recommendation algorithm that provides applicants the ability to make an informed decision regarding where to apply, as well as guides the decision-makers towards a more efficient process.

Graduation Semester

2016-08

Type of Resource

text

Permalink

http://hdl.handle.net/2142/92866

Copyright and License Information

Copyright 2016 Narender Gupta

Owning Collections