Stress is a state of worry and a natural response to a challenge, for students, it is often characterized by academic fatigue. It is a normal phenomenon everyone experiences, but it is excessive stress levels that lead to mental health problems, physical pain, bodily harm and even suicide. This study implements a highly efficient and accurate hybrid machine learning model to classify student stress levels, by taking into account the psychological, physiological, environmental, academic and social stress factors. Student stress level classification will enable early intervention, tailored support, evaluation of stress management programs, resource allocation and improvement of academic success. This also provides us with a deeper understanding of student stressors and how they are related. To build our hybrid model, we first implement five base models namely, Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR) and Extreme Gradient Boosting (XGBoost). We then combine these models and create the ensemble learning model that achieves an accuracy of 92.41%. The dataset includes 20 variables that contribute to students’ stress levels. We obtain our data from a Kaggle dataset which used a questionnaire for data collection. We improve the dataset balance using Synthetic Minority Oversampling Technique (SMOTE). Random Forest feature selection algorithm was utilized in extracting the variables with the highest feature importance in the feature space. To enhance the performance of the models, we also implement hyperparameter tuning and each model was evaluated using several performance metrics of accuracy, precision, recall and f1 score.