Analyzing the effect of difficulty level of a course on students performance prediction using data mining

Abstract

Recently the University Grants Commission of India has introduced a multistage examination system in higher education institutes in the country. The new system, called the Credit Based Continuous Evaluation and Grading System (CBCEGS), assesses a student on the basis of her continuous evaluation during the semester, combined with her performance in the end semester examination. This multistage examination pattern provides an opportunity to students to improve their performance. If a student cannot perform well in tests during the semester, she can improve her performance in the end semester test. But it does not seem so easy. In certain courses, due to their difficulty level such as mathematics, a student may not be able to improve her knowledge at the last moment despite hard work. Though, it may be possible in case of courses that are comparatively easy such as System Analysis and Design. This paper analyzes and predicts students performance using data mining techniques for two data sets of 1000 students each one for Mathematics, and the other for System Analysis, and Design. This study can help the education community to understand learning behavior of students as far as courses of varying difficulty are concerned. It is observed that Classification and Regression Tree (CART) supplemented by AdaBoost is the best classifiers for the prediction of students' grades for both subjects. J48 supplemented by AdaBoost performs excellent for System Analysis, and Design but perform worst for mathematics and M5P generates best results for early prediction of students' marks in the major test.