The paper presents a comprehensive approach to real time mental stress prediction using multi-modal physiological sensor data and explainable artificial intelligence (XAI) techniques. Neural network pruning techniques and quantization are implemented to optimize model efficiency for resource-constrained wearable devices while maintaining prediction accuracy. The key contribution of this work is the integration of Shapley additive explanations (SHAP) analysis to enhance model interpretability. The experimental results demonstrate that the proposed SHAP-guided pruned neural network model achieves a high accuracy of 96.27% and a weighted F1 score of 96.29% in distinguishing between baseline, stress, and amusement states, and offers transparent explanations for model predictions. This research advances the field of affective computing by combining effective real time stress detection with model optimization techniques and XAI methods, making the technology more trustworthy and applicable in real-world scenarios.