Machine learning-based prediction models are valuable prediction tools for assessing university performance as well as decision support tools for university governance and higher education system design. The prediction of student outcomes to enhance learning and teaching quality is one subject that has attracted considerable attention for different purposes. The first objective of this study is to develop and validate a prediction model using Machine Learning algorithms that predict students' outcomes in the case of Moroccan universities based only on the outcomes of courses taken in the previous semesters of university studies. This prediction model can be used as a basis for many subsequent studies on different aspects of higher education such as governance, pedagogy, etc. As a first application, we explore the responses of this prediction tool to analyze the outputs of the online learning experience that took place during the Covid-19 pandemic period. To achieve this, four machine learning algorithms are tested such as J48 decision tree, Random Forest, Multilayer Perceptron, and Naïve Bayes. The experimentations are developed by using Weka and the two metrics “accuracy” and “ROC Area” enable to assess the predictive performance of the models. The obtained results show that the Random Forest-based model provides superior results, as evidenced by its accuracy-ROC area, which reached an accuracy of 90% with a ROC Area of 95%. The use of this model to explore the outcomes of the distance learning experience taken during the Covid-19 pandemic, reveals a failure in the prediction performance of the model during the Covid-19 pandemic period, which indicates a change in the system's behavior during this period when teaching moved to the full online version in the year 2019/2020 and returned fully face-to-face in the 2021/2022 year. The failure in the machine learning algorithms' performance when the system changes its behavior can be a limitation of using prediction models based on machine learning in this context. On the other hand, these models can be used if they are properly designed to identify changes in the behavior of a system as shown in this study. Therefore, the proposed Random Forest-based model has the capability to forecast student outcomes accurately and can be applied for diverse analyses within the Moroccan education system. These analyses include but are not limited to identifying students at risks, guiding student orientation, assessing the influence of teaching approaches on student achievement, and evaluating training effectiveness, among others.

The use of machine learning algorithms for higher education performance assessment is an emerging area of research and several works have focused on student performance and related problems. The preliminary goal of this work is to determine and quantify the role of prerequisites in academic success by using machine learning algorithms with the Weka environment. The main objective is the development of a tool based on machine learning algorithms for the prediction of future results for a training program based solely on the previous academic profiles of the students. The interest is to link whether success in previous courses is associated with success in subsequent target courses. This will help to improve the planning of course sequences in a training program on the one hand and the overall academic students’ success on the other. The proposed methodology is applied for the analysis of the role of the prerequisites influencing courses success of a training course in Mathematical and Computer Sciences in a Moroccan university. For this purpose, we use several classification algorithms such as Random Forest, J48, and Multilayer Perceptron.
Preliminary results show that the correlation between the prerequisite reliability rates of the courses studied and the accuracy with which the learning algorithms predict the success outcomes of these courses is confirmed.
Also, these results show that the best accuracy and the best Receiver Operator Characteristic ROC area are obtained by using Random Forest algorithm and have reached 86% for the accuracy and 75.6% for the ROC area.