The aim of this article is to develop a machine learning model to predict the risk of academic failure among 320 first-year students in the Bachelor of Education program at ENS Tétouan. To achieve this goal, random forests were chosen for their ability to improve the robustness and accuracy of predictions. Our work is a process that begins with data collection and pre-processing, which includes cleaning, encoding and normalization steps. The data is then divided into training and test sets to evaluate model performance - the algorithm is trained on the training set, and its performance is measured using the following criteria: precision, recall and F-measure. Optimization and refinement of the model is achieved by tool algorithms scikit-learn has been used to implement and evaluate the algorithm, RandomForestClassifier, GridSearchCV, StandardScaler and the SMOTE technique which deals with class imbalance, thus improving model performance on imbalanced datasets are also implemented to refine the model, using scripts based on machine learning algorithms to visualize the graphs. Finally, we aim to provide an analytical approach to identifying students at risk of academic failure, enabling educators to implement early and targeted interventions, in order to proactively intervene to support struggling students.