Student Majors Prediction and Classification Using Machine Learning Techniques رسالة ماجستير

Abstract

Predicting students' majors is very important in developing the performance of the Palestinian Ministry of Education, initial recovery plans for weak students, determining the future needs of the Ministry of Education system, contributing to building structures for the teachers' organization within the Ministry of Education and many others, all of the above mentioned are the result of early anticipation of the students' different specializations. This study represents the first of its kind in Palestine. This thesis is concerned with determining the students' future specializations in the scientific, literary, industrial and commercial branches. The classification of majors is based on machine learning techniques and their classification ability. Identifying the essential features and the size of the dataset collected is very important to enhance classification accuracy and classification matrices. So, Finding and examining the best machine learning algorithms and their ability to classify as accurately as possible and at reach, high rates in different classification matrices depending on the selected feature and the collected data are the core of this thesis. The first step to achieving classification as accurately as possible is features selection. Specialists in the field of education determine these features. Those features included required information about the student, required information about the vi student's family, and other information about the student's community, in addition to his academic achievement in the tenth grade. This process is followed by creating a questionnaire judged by professionals, and then it is distributed to the various directorates of education after obtaining prior permission to do so. The collected dataset includes more than 1200 students in different branches; this dataset is subjected to the pre-processing process phase by cleaning, scaling, and encoding it to be ready for machine learning techniques. Sex classification learner algorithms are used, in addition to neural networks algorithms. The learner algorithms used are: k-nearest neighbors’ algorithm (KNN), Support vector machines (SVM), linear discriminant, decision tree, ensembled boosted tree, and ensembled bagged tree, while the neural network used is multiple-layer perceptron neural network (MLPNNs). Experiments show that neural networks significantly outperformed the rest of the other Machine learning techniques, as the classification accuracy reached 96.9%, Whereas the classification accuracy in ensembled algorithm id 93.8%, which is the best within the category of classification learner algorithms. The experiments also show poor accuracy when using marks in the curriculum alone for classification on all algorithm uses. This comparison shows the need to include other features that have been studied, the most important of these features are the student's and father's trends and the availability of specialization.