Abstract:

Higher education institutions aim to provide quality education to the students. One way to achieve this is by discovering knowledge for prediction about students’ performance. The knowledge is hidden among the data set and is extractable through knowledge discovery in data process. The present paper is designed to evaluate data mining process standard CRISP DM for the purpose of IT student’s performance prediction. In this paper, we construct data mining model that tries to predict student's academic success. Our data set consists of 401 students for three generations and we gathered information for 18 variables. Specific educational setting is used – university undergraduates and graduates in computer science.

This study has been carried out to answer following research question:
What are the differences in predictors of academic success between different generations of students?

The main question is further analyzed through two sub-questions: What variables are the best predictors of success?
Do student success predictors vary over time?
Research results indicated similarities between two generations (B and C) and differences between the “pair” (B and C) and generation A.

Among 15 analized factors three factors had similar results for all three generations, 8 factors were similar for the generation B and C, but were different for the generation A. Those eight were: lecture attendance, time management and learning time, score at the state graduation exam or admission exam, conscientiusness, personal learning space, which high school was finished previously, motivation type, working in teams. Of the 15 factors three were different for the “pair” (B and C): responsibility, seminar attendance, gender. One factor, the GPA (grade point average) in high school, had inconclusive results.

Keywords:

Academic success, success prediction, higher education, data mining, CRISP DM, neural network.