Developing a Predictive Model to Determine Higher Education Students’ Academic Status Using Data Mining Technology

Developing a Predictive Model to Determine Higher Education Students’ Academic Status Using Data Mining Technology

ABSTRACT

In contemporary higher education, the issue of student attrition has become a pervasive concern. To address the challenge of improving student retention, it is essential to delve into the intricate factors that underlie student attrition. In the realm of private higher institution education (PHIE), student attrition and retention are influenced by a multifaceted array of factors, spanning demographics, social dynamics, economic conditions, academic performance, and institutional aspects. These factors collectively contribute to the complex landscape of student attrition and retention in higher education. The primary aim of this research endeavor is to construct a predictive model, leveraging the power of data mining technology, to discern the likelihood of undergraduate students either persisting or leaving higher education institutions. This study meticulously follows a hybrid data mining process model, comprising six distinct steps: problem understanding, data comprehension, data preparation, data mining, knowledge evaluation, and knowledge application. For this investigation, guided by the understanding of the problem at hand, a selection of 15 relevant attributes is made, and a dataset of 7361 instances is employed to create and assess a predictive model with the capability to ascertain students' status. In the course of this study, various classification algorithms are harnessed in the model-building process. Notably, decision tree (J48), rule induction (PART and JRIP), and Bayes classifier (naïve Bayes) algorithms are deployed. To train and test the classifier model, a 10-fold cross-validation approach and a 66% split test option are utilized. Among the four algorithms subjected to testing, the decision tree classifier (J48) algorithm emerges as the most accurate, achieving an accuracy rate of 91.40%, followed by PART, JRIP, and naïve Bayes algorithms in descending order. Furthermore, the J48 algorithm reveals hidden patterns within the data, pinpointing certain factors as the predominant contributors to student attrition and retention. These include financial sources (self-sponsored, parent-sponsored, and scholarship), division (regular and extension), types of preparatory schools attended (private and public), department (computer science, accounting, marketing management, hotel and tourism, and management), educational background (social and natural sciences), and the year of completion of preparatory studies (before 1994EC-2001EC and after 2002EC-2009EC). It is worth noting that this study encountered challenges in data management, particularly in merging two table formats from the Student Record Management Information System (SRMIS). Additionally, the study highlights the importance of maintaining well-organized, accurate, and high-quality data in educational institutions to facilitate effective data analysis and insights.