Resumen en inglés

The Universidad de los Andes, like many universities in Latin America and around the world, faces a challenge with high rates of student dropout and an increase in the average number of semesters a student takes to complete their degree. Various research efforts aim to address this issue by developing tools for early detection of students at a high risk of failing a course. This approach could have a positive impact on the university by enabling early alerts and providing recommendations, suggestions, and support to students in such situations. Existing literature explores this problem from different perspectives, with a focus on implementing machine learning models to predict academic performance—essentially forecasting grades or categorizing them into ranges. However, most of these studies concentrate on the outcomes of these models rather than exploring the broader implications of the results. This research will analyze the most common machine learning models discussed in the literature for addressing student performance issues, such as Logistic Regression (LR), Support Vector Machine (SVM), Multilayer Perceptron (MLP), Random Forest (RF), Gradient Boosting (GBoosting), XGBoost (XGB), and Long-Short Term Memory (LSTM). Various combinations of hyperparameters will be considered for different training representations. Subsequently, using the results, a Knowledge Tracing (KT) model based on concepts will be developed. This model aims to provide the university with information about potential concepts or knowledge areas that need reinforcement and pose a risk to the student. This approach will enhance the prediction model by considering a stu dent’s specific concepts for a particular subject