Predictive Modeling of Student Dropout Using Academic Data and Machine Learning Techniques

Abstract

This study's objective is to investigate the performance of a predictive model for students at risk of dropout (DO) by considering several internal criteria of an academic program. This research uses academic information from UIN Syarif Hidayatullah Jakarta and applies the C4.5, Naive Bayes Classification (NBC), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) to forecast which students might drop out. The data used consists of 714 student records from Department of Information Systems for the academic year 2010–2015 as training and 2018 as testing data. The research method refers to the SEMMA framework (Sample, Explore, Modify, Model, and Assess) to ensure systematic and accurate data processing. Meanwhile, the internal criteria used are the completed courses, the status of the internship report, and the final project proposal. According to the study's findings, the C4.5 and SVM algorithms get the best accuracy rates of 94.44%, while KNN and NBC come in second and third, respectively, with 93%. The results show that the C4.5 and SVM algorithms work well with academic data. This study provides a substantial contribution to the development of a prediction system for students at risk of dropping out, which can be integrated into data-based applications or dashboards. This solution is expected to help higher education institutions identify students who need further academic support. In addition, this research also opens up opportunities for the progress of more accurate forecasting models through the integration of additional variables such as behavioral or psychological data. With this data-driven approach, higher education institutions can enhance their efficiency in monitoring and preventing student dropouts, thereby supporting a vision of quality and sustainable education.