The tremendous growth in electronic data of universities creates the need to have some meaningful information extracted from these large volumes of data. The advancement in data mining field makes it possible to mine educational data for improving the quality of the educational processes. This dissertation, thus, uses data mining methods to study the performance of undergraduate students. Two aspects of students’ performance have been focused on. Firstly, predicting students’ academic achievement at the end of a 4-year study programme, and secondly, studying typical progressions and combining them with prediction results. Predicting performance of students at the end of a university degree at an early stage of the degree program would help universities not only to focus more on bright students but also to initially identify students with low academic achievement and find ways to support them. The data of four academic cohorts of three faculties at NED University of Engineering & Technology, comprising 347 undergraduate students of Computer Science and Information Technology, 587 undergraduate students of Civil Engineering and 430 undergraduate students of Electronic Engineering, have been mined with different classifier models. The results show that it is possible to predict the graduation performance in final year at university using only pre-university marks and marks of first and second year courses, no socio-economic or demographic features, with a reasonable accuracy. Using only marks for students’ performance prediction and no other socio-demographic features will enable university administration to develop an educational policy that is easier to implement. This is the reason to investigate whether acceptable results can be obtained with marks only. Further, data of one cohort of students are used to predict students’ performance of the following cohort to test the generalizability and therefore the actionability of our approach. Moreover, using these classifiers, we explore how to derive courses that can serve as effective indicators for students’ performance at an early stage of the degree program for timely intervention. Indeed, once such courses are put in evidence, performance of students at the end of a course could be predicted and would allow for intervention while the indicator courses are actually taking place. A pragmatic policy is proposed to derive those indicators based on decision trees, a kind of classifiers that is explained in Chapter 2, Section 2.1.3.1. As the obtained decision trees have a lower accuracy than two other classifiers, though it is still acceptable, the goodness of the pragmatic policy needs to be further investigated. Therefore, we investigate how academic performance of students evolves over the four-year degree as a kind of triangulation. For this purpose, students of two consecutive cohorts of Computer Science and Information Technology have been clustered each year taking their final examination marks in individual courses in each of the four years. X-means and K-means clustering taking Euclidean distance for both algorithms have been applied. We put in evidence interesting typical progressions in particular students who have low marks all the way through their studies and students with high marks throughout their studies. The key contribution of our work is to understand the benefits of the pragmatic policy that is proposed earlier in this work. It turns that our pragmatic policy uncovers (almost) all the targeted students: students with low marks and students with high marks. Therefore, its implementation can be recommended.