Stress is a tremendous impact on human health, often leading to psychological and physiological diseases if not diagnosed or controlled. This work provides an intelligent human stress detection system based on a Peafowl-Optimized XGBoost classifier that achieves good accuracy and reliability in stress recognition. The system takes physiological signal data as input and applies Gaussian filtering and Z-score normalization during preprocessing to remove noise and normalize characteristics. Subsequently, entropy-based feature extraction is used to capture complicated signal fluctuations associated with stress levels. The collected features are then classified using an XGBoost model improved by the Peafowl algorithm, which improves the model's convergence speed and classification accuracy. Experimental results show that proposed model achieves 98% accuracy, precision, recall and F1 score with overall classification accuracy of 98.5% confirming its effectiveness for accurate stress level detection.