Artificial intelligence has established itself as a transformative tool in higher education, revolutionizing the paradigms of analysis and improvement of academic performance in virtual learning environments. This study conducts a systematic literature review on artificial intelligence applications for predicting academic performance in virtual education, employing the PRISMA methodology. An exhaustive search was carried out in recognized academic databases, such as Scopus, ScienceDirect, SpringerLink, IEEE Xplore, and Web of Science. Inclusion and exclusion criteria were established to select relevant studies, resulting in a total of 762 studies, of which 41 were selected after applying the defined criteria. The analysis reveals a predominance of techniques such as random forest (25%), support vector machines (18%), decision trees and logistic regression (14%), and k-nearest neighbors (11%). The most significant predictor variables include: historical academic performance (34%), interaction patterns on platforms (29%), demographic data (23%), participation in activities (8%), and emotional states (3%). As future work, a multivariable predictive model is planned to be developed, integrating emotional and contextual dimensions using hybrid deep learning architectures.