Mental stress, a typical response to everyday difficulties, may have serious adverse effects on one’s physical and psychological health. This work explores the classification of stress levels from EEG data using machine learning (ML) and Deep Learning (DL) approaches. The EEG data are first processed to extract 28 features that capture important patterns of brain activity associated with stress. Three feature selection models, Random Forest (RF), Support Vector Machine (SVM), and XGBoost (XGB), are used to choose these features. Several classification models, including XGB, RF, SVM, and Long Short-Term Memory (LSTM), are then trained using the chosen features. The findings demonstrate that a remarkable 99.67% accuracy in stress level classification is achievable when combining the RF feature selection model and the LSTM classification model. This study demonstrates the efficacy of ML-based feature selection in conjunction with DL models for real-time stress detection, emphasizing its applicability in mental health monitoring and tailored stress treatment.