In the context of increasingly competitive university admissions and the growing need for personalised academic planning, this study proposes an intelligent two-stage recommendation model for major selection. The first stage of the framework utilises Holland’s RIASEC model to evaluate students’ personal interests and vocational orientations. The second stage leverages machine learning algorithms to estimate the probability of admission based on entrance exam results, subject combinations, and student preferences. Each academic major is associated with a predefined RIASEC vector, and its similarity to the student’s interest profile is quantified using cosine similarity. The applicant is prompted to enter relevant data. A composite score is then derived by weighting both the admission probability and the interest alignment, helping students select majors that match their profile while maintaining a high likelihood of admission. The effectiveness of the proposed model was validated using a dataset of 5,503 students with 16,326 application preferences to Can Tho University of Technology for the 2024 academic year. The results show that the Stacking model achieved 88.76% accuracy and demonstrated robust performance on the test set, confirming its predictive reliability. Additionally, when applied to simulated data, the model proved to be feasible and highly aligned with student preferences, indicating strong potential for real-world implementation.