Ramalan Prestasi Akhir Pelajar Melalui Kaedah Perlombongan Data [Student Final Performance Forecasting Through Data Mining Methods]

Abstract

Cabaran besar yang dihadapi oleh pensyarah dalam merancang dan melaksanakan pengajaran dan pembelajaran (PdP) berkesan adalah mengenalpasti pelajar yang berprestasi lemah pada peringkat lebih awal sebelum mereka bergraduat. Bagi menyahut cabaran ini, kajian ini dibangunkan bertujuan untuk meramal prestasi pelajar yang menyambung pengajian di Kolej Komuniti Kuala Langat berdasarkan keputusan kemasukan Sijil Pelajaran Malaysia (SPM) dengan menggunakan kaedah perlombongan data melalui teknik pengklasifikasian. Empat (4) algorithma pengklasifikasian telah digunakan untuk membina model ramalan keputusan akhir peperiksaan melalui Himpunan Purata Nilaian Mata (HPNM) iaitu  (1) Pepohon Keputusan (J48), Multilayer Perceptron (MP), Hutan Rawak (RF) dan ZeroR. Set data yang digunakan di dalam kajian ini diperolehi daripada Sistem ePelajar dan Integrated Student Evaluation Management System (iSEMS), Jabatan Pengajian Politeknik dan Kolej Komuniti (JPPKK). Keputusan kajian menunjukkan bahawa model pengklasifikasian J48 adalah model terbaik dengan nilai ketepatan 56.56% diikuti oleh algoritma ZeroR dan RF, kedua-dua dengan nilai ketepatan 50.67%. Manakala algorithma MP memperolehi ketepatan terendah dengan peratusan 44.34% sahaja. Selain daripada itu, didapati matapelajaran Bahasa Melayu (BM) dan Sejarah (SJ) mempunyai pengaruh terhadap pengkelasan status HPNM pelajar. Seramai 49/76 orang pelajar telah dikenalpasti mempunyai kebarangkalian HPNM yang rendah di semester akhir. Kajian ini boleh membantu institusi dalam mengenalpasti dari semester awal pelajar yang diramalkan berprestasi rendah supaya program bersasar boleh dianjurkan untuk melonjakkan prestasi pelajar.

The big challenge faced by lecturers in planning and implementing effective teaching and learning (PdP) is to identify students who perform poorly at an earlier stage before they graduate. To meet this challenge, this study was developed to predict the performance of students who continue their studies at Kuala Langat Community College based on the results of admission to Sijil Pelajaran Malaysia (SPM) by using data mining methods through classification techniques. Four (4) classification algorithms have been used to construct the final results prediction model of the examination through the Asset Points Assessment System (HPNM) namely (1) Result Tree (J48), Multilayer Perceptron (MP), Random Forest (RF) and ZeroR. The data set used in this study was obtained from the ePelajar System and Integrated Student Evaluation Management System (iSEMS), Department of Polytechnic Studies and Community Colleges (JPPKK). The results of the study show that the J48 classification model is the best model with an accuracy value of 56.56% followed by ZeroR and RF algorithms, both with an accuracy value of 50.67%. While the MP algorithm obtains the lowest accuracy with a percentage of only 44.34%. In addition, it was found Bahasa Melayu (BM) and history (SJ) has an influence on the classification status CGPA students. A total of 49/76 students were identified as having low probability of HPNM in the final semester. This study can assist institutions in identifying from the initial semester of students who are predicted to perform low so that targeted programs can be organized to boost student performance..