A Prediction and Visual Analysis Method for Graduation Destination of Undergraduates Based on LambdaMART Model

Introduction

Graduation is an important turning point in a student's life. After graduation, one typically has five destinations, including domestic graduate school, overseas study, employment, freelance work, and unemployment. The likelihood of embarking on a particular destination is influenced by various factors that include gender, birthplace, academic performance, and personal strengths (Zhang et al., 2021). Using an individual’s circumstances to predict a student’s graduation destinations is important; for students, it helps them clarify their future direction and make reasonable study plans in advance; for teachers, they can optimize their curriculum and provide career guidance to students (Peng, 2020; Liu, 2021).

Generally, the length of undergraduate study is four years, or eight semesters, and about 60 subjects are required. Each subject has credit requirements and assessment results. Each undergraduate accumulates a large amount of data from enrollment to graduation, which includes basic information (including gender, birthdate, birthplace, political affiliation, and major), course grades, graduation destinations (including destination type and industry type), etc. This information constitutes a huge undergraduate academic data set. How to efficiently and deeply analyze these data, find out the distribution characteristics of course grades and graduation destinations, explore the factors that affect students' graduation destinations, and then accurately predict students' graduation destinations is quite challenging.

Most of the existing analysis methods are based on questionnaires, statistical analysis and other methods. Although better results have been achieved, it is difficult to analyze the data in depth in all aspects (Tawafak et al., 2019). The visual analytic techniques that have been developed in recent years map complex data into easily perceivable graphical, symbolic, color-coded and other representations. They are also supplemented with interactive means to enhance people's ability and efficiency in analyzing data so as to quickly discover the features and patterns hidden within the data and provide new ideas for intuitive in-depth analysis of academic data (Alkhalil et al., 2021).

Most existing data mining methods (such as association rule mining, decision trees and classification) and machine learning methods (such as logistic regression) have been used for undergraduate graduation prediction (Gulzar et al., 2019; Jaiswal et al., 2019). This has achieved better results, but the prediction accuracy still needs to be improved. A new machine learning model called LambdaMART that was proposed in recent years can solve the ranking problem directly and effectively without performing classification or regression. This provides a new solution for accurately predicting the graduation destination of undergraduate students.

In this paper, we propose a prediction and visual analysis method for graduation destination of undergraduates based on LambdaMART, called PGDU_LM, and a system called PGDUvis that helps students and teachers deeply analyze the correlation between each subject and the graduation destination, as well as predict the graduation destination based on course grades. The main contributions of this paper can be summarized as follows:

• 1.

  A prediction algorithm for the graduation destination of an undergraduate, called PGDU_LM, is proposed; the algorithm uses Spearman correlation coefficient to analyze the correlation between subjects and graduation destinations, extracts characteristic subjects, and uses the LambdaMART ranking model to calculate students' propensity scores in different graduation destinations and then predicts their graduation destinations. The prediction accuracy can reach 86%.

• 2.

  A visual analysis method of the course grades and graduation destinations of students is designed. It supports users to analyze student data from multiple dimensions, such as students' course grades, birthplace, graduation destination, industry type, gender, etc., and explore the factors that influence the graduation destination of students.

• 3.

  A prediction and visual analysis system for graduate destination of undergraduates called PGDUvis is designed and implemented. It supports users to interactively analyze the academic data of graduates and predict the graduation destination of undergraduates.