The thesis covers the application and evaluation of classification algorithms to predict failures and potential academic performance of students based on Portuguese secondary school students data.
The application of Educational Data Mining methods on the database of Portuguese secondary school students is promising to ascertain the factors having a dominant influence on student's academic performance and to identify students who has a high chance to fail from subject or dropout. The binary and multiclass classification approach is followed. The comparative analysis of the performance results of the four classification algorithms is covered. As performance metrics accuracy and precision are selected. The prediction results are highly influenced by first and second-period grades, when these features are discarded, all classifi cation algorithms cannot generalize the data. For future research, more features related to student socio-economical and academic background can be added, in order to achieve highly accurate classi fication solutions which will be more powerful at extracting useful patterns about student's academic performance and predicting the academic future of student more precise.