Multimodal learning analytics (MMLA) has become a prominent approach for capturing the complexity of learning by integrating diverse data sources such as video, audio, physiological signals, and digital interactions. This comprehensive review synthesises findings from 177 peer-reviewed studies to examine the foundations, methodologies, tools, and applications of MMLA in education. It provides a detailed analysis of data collection modalities, feature extraction pipelines, modelling techniques—including machine learning, deep learning, and fusion strategies—and software frameworks used across various educational settings. Applications are categorised by pedagogical goals, including engagement monitoring, collaborative learning, simulation-based environments, and inclusive education. The review identifies key challenges, such as data synchronisation, model interpretability, ethical concerns, and scalability barriers. It concludes by outlining future research directions, with emphasis on real-world deployment, longitudinal studies, explainable artificial intelligence, emerging modalities, and cross-cultural validation. This work aims to consolidate current knowledge, address gaps in practice, and offer practical guidance for researchers and practitioners advancing multimodal approaches in education.

Los problemas de bajo rendimiento académico y rezago son recurrentes en instituciones educativas de nivel superior, especialmente al inicio de los estudios universitarios. En el contexto local, análisis diagnósticos han mostrado altos índices de reprobación y bajo rendimiento académico. En este trabajo, se utilizaron datos sociodemográficos y resultados de exámenes de admisión de 415 alumnos de las carreras del área de computación de la Universidad Autónoma de Yucatán (México), inscritos entre 2016 y 2019. El objetivo es generar modelos predictivos de riesgo académico, empleando métodos de la minería de datos educativa, que sirvan como herramientas de detección temprana de condiciones de riesgo académico y faciliten el despliegue de estrategias de intervención educativa. Se siguieron las etapas del Proceso de Extracción de Conocimiento en Bases de Datos, concretamente, se aplicaron técnicas de clasificación para el análisis, obtención y validación de los modelos. Los resultados muestran que el mejor modelo corresponde al algoritmo LMT, con un valor de precisión de 75.42% y un 0.805 para el área bajo la curva ROC. Se logró identificar a los mejores atributos predictores, particularmente las pruebas del examen de ingreso a licenciatura fueron muy significativas. Se propone el desarrollo de herramientas informáticas para la detección precoz de riesgo académico y estrategias de intervención educativa oportuna. The problems of poor academic performance and lag are recurrent in higher-level educational institutions, especially at the beginning of university studies. The early detection of academic risk conditions enables the implementation of educational intervention measures to address factors of poor school performance, associated with the particular contexts of the students. The purpose of this study was to generate predictive models of academic risk, using educational data mining methods, specifically classification or prediction techniques, for the analysis, obtaining and validation of the models. The data used correspond to admission exam results and sociodemographic data of 415 students of the computer science majors at the Autonomous University of Yucatán (Mexico), enrolled between 2016 and 2019. The results show that the best model corresponding to the algorithm of LMT classification, with a precision value of 75.42% and 0.805 for the area under the ROC curve. It was possible to identify the best predictive attributes, particularly the bachelor entrance exam tests were very significant. The development of computer tools for the early detection of academic risk and strategies for timely educational intervention is proposed.