2026-10-12T10:17:21Z https://www.scidb.cn/oai

10.57760/sciencedb.j00104.00145 2026-09-10T10:53:38Z

2026-09-10 Education doi:10.57760/sciencedb.j00104.00145 en We use three real-world datasets, HNMU, VNU, and SATDAP, to evaluate the performance of the proposed model. The HNMU and VNU datasets contain information on students' academic performance and survey feedback from two major universities in Vietnam. Each dataset differs in scale, feature distribution, and classification labels, creating a diverse testing environment for our approach. The inclusion of the SATDAP dataset, collected in Portugal, enhances the generalizability of our proposed method by incorporating data from a distinct international educational context.The classification labels used in this paper are determined based on real-world data from the three datasets: HNMU (collected from Hanoi University of Science and Technology), VNU (collected from Vietnam National University), and SATDAP (collected from the SATDAP - Capacitação da Administração Pública program under the POCI-05-5762-FSE-000191 project in Portugal). Among these, the HNMU dataset has four classification labels: Average, Good, Very Good and Excellent; while the VNU dataset has three labels: Good, Very Good and Excellent. Since both universities only store data for students who have graduated and do not have information on students who did not graduate, the dataset does not include the "Fail" label. The SATDAP dataset has three classification labels: Graduate, Enrolled, and Dropout.The HNMU dataset comprises 551 samples with 86 attributes collected from students enrolled in Mathematics Education at HNMU. The attributes in this dataset include results from specialized courses, the number of credits completed, and evaluations from instructors.Similarly, the VNU dataset includes data from 271 students studying Literature Education at the University of Education (VNU) spanning from 2014 to the present and comprises 92 attributes. The data reflects academic performance each semester, participation in scientific research activities, and some personal characteristics. The collection and preprocessing of these datasets spanned three years, from 2021 to 2023.The SATDAP dataset contains 4,424 samples with 36 features, covering the learning journey, demographic information, socio-economic factors, and student grades over a five-year period. The dataset includes variables related to demographic factors (age at enrollment, gender, marital status, nationality, postal code, disability status), socio-economic factors (employment status, parents' education level, parents' occupation, parents' employment status, student scholarships, student debt), and the learning journey (entrance scores, number of years repeating in high school, field of study choice order, and type of high school program).To ensure data consistency and minimize bias, we applied preprocessing steps such as data cleaning, normalization, and feature extraction. Additionally, synthetic data generation techniques were used to increase dataset stability and improve the model's generalization ability. Graduation Classification; Limited Training Samples; training dataset Nguyen Thi Kim Son PUBLIC https://creativecommons.org/publicdomain/zero/1.0/ dataset Science Data Bank