Increasing productivity in educational systems is of great importance. Researchers are keen to predict the academic performance of students; this is done to enhance the overall productivity of educational system by effectively identifying students whose performance is below average. This universal concern has been combined with data science leading to the creation of an interdisciplinary research area called Educational Data Mining. One of the recent issues which has been addressed by researchers is training generalizable models from different aspects such as gender, major, geography and etc. Therefore, in this research we use machine learning methods to predict student's performance, emphasizing on training generalizable models from geographical aspect. For this purpose, a questionnaire containing 37 questions was designed, through which 536 answers were collected, including 111 international and 425 domestic answers. According to the literature, student performance is mostly determined based on the GPA (grade point average) of the entire course. In this research, information about the GPA of respondents in undergraduate and graduate courses was collected in the form of three classes. After a final review of the models employed in previous studies, the main models selected and used for classification purposes included SVM, CNN, Adaboost, RF, SVM, and DT. Feature selection is performed using XGBoost, random forest, as well as SVML1. The main issue investigated in this study is the generalizability of the models trained on domestic (Iranian) data and tested on international data (non-Iranian). Experimental results show that the best models trained with specific dataset collected in this research had generalizability comparing to base models' outcomes which were trained and tested on domestic data. Meanwhile, Random forest and CNN models shows the best performance with the average of accuracy and F-score of 73.5 and 68.5, respectively.