STUDENT ACADEMIC PERFORMANCE PREDICTION FRAMEWORK WITH FEATURE SELECTION AND IMBALANCED DATA HANDLING

• Published in: Jurnal Ilmiah Kursor: Menuju Solusi Teknologi Informasi Vol. 12; no. 3; pp. 123 - 134
• Main Authors: Wijayaningrum, Vivi Nur, Kirana, Annisa Puspa, Putri, Ika Kusumaning
• Format: Journal Article
• Language: English
• Published: Informatics Department, Engineering Faculty 25.05.2024
• Subjects: classification; drop out; Random Forest; SMOTE
• ISSN: 0216-0544; 2301-6914
• DOI: 10.21107/kursor.v12i3.356

Various factors cause the low scores of students in practicum courses. If these factors cannot be identified, more and more students will drop out of the study due to low scores, especially Vocational College students who do not have the opportunity to improve their scores in the short semester. Students with the potential to drop out must be identified as soon as possible because the number of dropouts can have an impact on a university's accreditation value. In this study, the prediction of student academic performance was carried out using a framework consisting of imbalanced data handling using SMOTE and feature selection using Random Forest, as well as the application of Multi-Layer Perceptron (MLP) for the formation of a classification model. The MLP architecture consists of some neurons in the input layer, two hidden layers with five neurons each, and two neurons in the output layer. SMOTE succeeded in selecting ten significant parameters from 22 initial parameters, which produced the most accurate predictions. According to the test results, the proposed framework offers the best accuracy of 0.8889 and an F1-Score of 0.9032. These results prove that the proposed framework can be used as an alternative for the Department to take action to prevent students from dropping out.