Abstract
In higher education, especially universities implementing their education according to the credit system, each student needs to have a study plan suitable for themselves to save time, costs and achieve the best learning results. To do this, it is necessary to closely monitor students' learning progress with support tools. These tools mine student score databases to forecast their learning performance as early as possible, thus helping managers, instructors and advisors to guide students' learning plans, identify special cases and provide appropriate support. This thesis propose a method to forecast student's academic performance using a deep learning model named Multivariate Long-Short Term Memory (MLSTM). The score database of students are collected, analyzed, pre-processed, chosen to create input data of MLSTM training phase. In addition, we experiment forecast models using Long-Short Term Memory (LSTM), Multi-Layer Perceptron (MLP), and Support Vector Regression (SVR) to compare the effectiveness of the proposed model. The experimental dataset is acquired from student's learning results of Vinh Long University of Technology Education with 367,351 records according to 10,208 students. Through experimental results, the proposed MLSTM model gives the most accurate prediction results with a prediction error - calculated by the RMSE (Root-Mean-Square Error) formula of 0.7, completely feasible for practical application.