Predicting Student’s Performance Based on Behavior and Online Engagement Activities Using Machine Learning رسالة ماجستير

Abstract

Predicting students’ performance is important in improving students’ outcomes in Higher Educational Institutions, especially after the spread of educational platforms as a result of Corona epidemic. Students’ behavior data and engagement activities recorded in Learning Management Systems could be used in predicting students’ performance to help in determining the students’ needs, which help students and university. In addition, it contributes to set development plans using ML models to serve this field. The thesis is concerned with determining the benefits of using educational platforms especially Moodle through gathering students’ attributes such as behavioral data and engagement activities to predict student’s performance. This is somehow problematic in many ways like extracting the data then deciding which students’ features to use in the study and which students’ features are more beneficial in predicting students’ performance. The problem of the study also appears in designing the framework, choosing the machine learning techniques, testing the algorithms and selecting only one that suits the study. The data were subjected to pre-processing operations in order to get rid of the random partition, empty value processors, deleting duplicate data, describing, and coding the data. After that, these data were subjected after processing operations to different machine learning techniques as well as neural network techniques, where four techniques were used for classification according to an associated set of MATLABsupported algorithms: KNN, TREE, Ensembled Tree and SVM. vi This thesis reached several results. The most important of which are: The SVM technique was superior to other classification techniques, as the classification accuracy in this technique reached 90%, while the classification value in KNN was equal to 87%. MLPNNs neural networks gave better results, the accuracy of classification was 93%, neural networks were able to classify and predict with high accuracy, this leads us to many studies on the inclusion of other techniques of neural networks in the future.