A Two-Stage Early Prediction Model to Monitor the Students' Academic Progress

The high dropout rate and the low percentage of undergraduate students who graduate on time are some of the problems at higher education institutions. Various research has been conducted to overcome these problems, one of which is predicting student performance. Predicted results of academic performance at the beginning of the class can be used as an early warning for undergraduate students and lecturers to take the necessary actions to succeed well at the end of the semester. However, not all significant predictor variables can be obtained at the beginning of the class. The purpose of this research is to develop a two-stage prediction model to determine the passing of each student from the courses undertaken. The first stage of the prediction model was developed at the beginning of the class while the second was developed at the end of the eighth week by adding two predictor variables, namely mid-term test score and the number of presence in class. Four different methods are used: Decision Tree (able to generate rule that are easy to interpret), Random Forest (able to handle high dimensional data), Support Vector Machine (able to eliminate overfitting problem), and Logistic Regression (able to make good prediction of success in a course. In testing, all performance measures of Logistic Regression method were superior to other methods both in the first and second stages. It is also seen that the addition of predictor variables was able to increase the performance measures of all prediction models by up to 7%.