Impact of Different Factors on the Selection of the Degree Stream of University Students

Abstract

Academic failure among university students studying in universities and higher educational institutes has become a major problem since recently. Numerous researches are being done to understand the reasons for this situation. Among these researches, Educational Data Mining (EDM) has played a major role in the past few years. One reason for these failures is that the students do not have a clear understanding about the degree programs when they enter the institute and have no idea about the knowledge offered through these different degrees. Sometimes students tend to select degree programs on someone else’s opinion too. This research aims to identify the factors such as past educational results, main subjects followed in previous studies, success in entry level module exams and any other potential factors that may have an effect on the performance of the university students and hence affect their selection of the degree program. In this research, information about students of Sri Lanka Institute of Information Technology (SLIIT) has been used to understand these factors. Among many degrees available at SLIIT, only three main degree areas are considered for this project: Bachelor of Science (Special Honours) in Information Technology, Bachelor of Science (Special Honours) in Computer Systems & Networking, and Bachelor of Science (Special Honours) in Information Systems. This thesis describes different data mining methodologies used to identify performance factors and provide the most significant factors correlated to academic success. First an initial analysis of the data set was done to understand the data set and based on the nature of the data set it was decided to classify students into two groups as ‘fail’ and ‘pass’ according to their performance which is measured using the Cumulative Grade Point Average (CGPA). In this study data mining techniques such as Neural Networks and Binary Logistic Regression are used. Generated models are evaluated to identify the best fitting model for the data set using classification tables, gain charts, lift charts and ROC curves using Area Under the Curve (AUC) values. These results and models can later be used to predict the most suitable degree stream for a new student in order to successfully complete the degree and also it will be useful for institutions to plan the programs accordingly, provide guidance for students, setting standards and in turn produce students with great success.