A Model Proposal for Predicting Students’ Academic Performances Based on Data Mining

A Model Proposal for Predicting Students’ Academic Performances Based on Data Mining

Murat ALTUN, Kemal KAYIKÇI & Sezgin IRMAK

ÖZET
The purpose of this research is to propose a model to predict the academic performances of students and prevent their possible academic failures in the future. This research was conducted as an educational data mining application. The academic data of 1570 students who graduated from four departments of Akdeniz University, Faculty of Education between the year 2012 and 2017 were used in the research. All the exam scores, final grades, weighted averages of semester grades, and graduation grades were used in the study. Two main models have been developed for predicting students' academic success using data mining techniques and algorithms. The First Model is the Student Graduation Grade Estimation Model. This model is aimed to predict the future graduation grades of the students. Sub-models were developed using Artificial Neural Networks and Multiple Linear Regression Analysis. It was observed that the developed models predict the graduation grade of the students with an accuracy of 94% to 97% from the 1st semester’s data. The second model developed in this research is the Early Warning Model for Students’ Possible Academic Failures in the Future. The model predicts whether the general weighted average grades will fall below 2 in the future, according to the students’ 1st year’s 1st-semester grades. Under this model, the accuracy of the sub-models which were developed using Logistic Regression and Decision Trees was found to be 72% to 87%. As a result of the research, a model was proposed to prevent the academic failures in the future by predicting the student's academic performances. It can be asserted that educational institutions can benefit effectively and efficiently to increase students success by using the proposed model.

ABSTRACT
The purpose of this research is to propose a model to predict the academic performances of students and prevent their possible academic failures in the future. This research was conducted as an educational data mining application. The academic data of 1570 students who graduated from four departments of Akdeniz University, Faculty of Education between the year 2012 and 2017 were used in the research. All the exam scores, final grades, weighted averages of semester grades, and graduation grades were used in the study. Two main models have been developed for predicting students' academic success using data mining techniques and algorithms. The First Model is the Student Graduation Grade Estimation Model. This model is aimed to predict the future graduation grades of the students. Sub-models were developed using Artificial Neural Networks and Multiple Linear Regression Analysis. It was observed that the developed models predict the graduation grade of the students with an accuracy of 94% to 97% from the 1st semester’s data. The second model developed in this research is the Early Warning Model for Students’ Possible Academic Failures in the Future. The model predicts whether the general weighted average grades will fall below 2 in the future, according to the students’ 1st year’s 1st-semester grades. Under this model, the accuracy of the sub-models which were developed using Logistic Regression and Decision Trees was found to be 72% to 87%. As a result of the research, a model was proposed to prevent the academic failures in the future by predicting the student's academic performances. It can be asserted that educational institutions can benefit effectively and efficiently to increase students success by using the proposed model.

ANAHTAR KELİMELER: Prediction of the academic performance, academic warning system, educational data mining, decision trees, artificial neural networks

KEYWORDS: Prediction of the academic performance, academic warning system, educational data mining, decision trees, artificial neural networks

DOI : [PDF]