Intelligent Model for Suitable University Specialization Selection in Palestine

Abstract

Choosing suitable track is a key success in the academic and professional life. Whenever the specialization is appropriate for the student; an increase in students' performance is the natural result. In addition, it is an impor- tant theme in a persons life and it is one of the vital topics which deserve to be searched. Many studies investigated the in uential factors a ecting spe- cialization selection by using statistical methods, but none of the researches studied these factors and employed machine learning methods to a develop classi cation model which can help to choose a suitable specialization. In this research, we extracted the local in uential factors in our area (Palestine) by using lter approach Correlation-based Feature Selection (CFS) and factor analysis approach Principle Component Analysis (PCA). Accord- ing to the results, we identi ed ve basic in uential factors a ecting special- ization selection at the university in Palestine, which is Personal factors, de- mographic factor, university factor, family background factor and academic achievement factor. Then we developed a classi cation model which might consider the rst proposed model studying the in uential factors a ecting the specialization selection and has the ability to predict the specialization selection for high school students by identifying the suitable specialization based on rules. To accomplish this; an extensive literature review was con- ducted. We identi ed the in uential factors a ecting the specialization selec- tion based on the literature review, such as students grades, parents in uence, suitability for students talents and nancial conditions and other factors as well. A special questionnaire was developed which covers various questions re- lating the in uential factors. The questionnaire was tested and arbitrated by 10 experts. Hence, our proposed model depends on extracting the previous knowledge and student experiments. The collected data used as inputs to build our classi cation model using PART. PART classi er is a rule-based classi er that is built on partial decision tree without the need for global optimization. The method combined between C4.5 method to generate a decision tree from the training data and RIPPER approach used to gener- ate rules by two stages which are repeated growing stage and pruning stage. After developing the classi cation model, it was tested on a set which was not used in the training model. It contains 70 student instance, who do not know their class (major), but they want to know what future specialization ts them. After they answered the questions, the classi cation model can use to predict the suitable major for each student. According to the results, the researcher found that the accuracy of the proposed model is 77.4% for the training group. And for the testing group, the results showed that the accuracy of the proposed model was 73.7%. So the developed model adopted nal 49 rules which considered as a map to lead high school students steps toward choosing the suitable specialization.