Depression is a high-priority public health issue in universities because it affects students’ emotional well-being and academic performance. The objective of this study was to implement a hybrid model for classifying depression based on tabular data from standardized surveys (USDI-30). The proposed methodology was developed in five phases: dataset acquisition, preprocessing, feature extraction (TabNet, FNN, DCN, and FT-Transformer), implementation with machine learning techniques (logistic regression, SVM, random forest, and XGBoost), and evaluation using stratified cross-validation and hyperparameters to maximize performance. The superior results were obtained with the FT-Transformer + XGBoost model with the metrics accuracy, precision, recall and F1_macro, whose results were 95.12%, 94.52%, 96.12%, and 95.24%, respectively. Overall, the proposed model integrates deep representations and robust classifiers, providing an effective tool for the early detection of depression and support for mental health decision-making among students.